Adaptive 2D Posture Analysis

Adaptive 2D Posture Analysis | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Adaptive 2D Posture Analysis Hemanth Kumar Budige This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6657445/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Poor posture is a leading cause of musculoskeletal disorders and reduced biomechanical efficiency in various populations, including office workers, athletes, and individuals undergoing physiotherapy. This research presents an adaptive 2D posture correction system based on real-time human pose estimation and dynamic feedback mechanisms. Utilizing lightweight 2D pose estimation frameworks such as MediaPipe and OpenPose, the system captures key body landmarks and calculates posture deviations based on predefined ergonomic baselines. A novel keypoint augmentation strategy introduces a synthetic "neck" keypoint between the nose and shoulder midpoints, enhancing posture evaluation accuracy. Furthermore, adaptive feedback through visual and auditory cues enables immediate correction in environments such as fitness, yoga, and rehabilitation. The model demonstrates robust performance under diverse lighting conditions, occlusions, and clothing variances. Results indicate a 92.4% improvement in posture classification accuracy using the proposed keypoint augmentation. This work paves the way for intelligent, accessible, and real-time posture improvement applications. Software Engineering Pose Estimation Posture Correction Machine Learning Real-time Feedback Ergonomics Computer Vision Human Activity Recognition MediaPipe OpenPose Rehabilitation Physiotherapy Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction The widespread adoption of sedentary lifestyles and increased reliance on digital devices have led to a surge in posture-related health issues. Improper posture, sustained over prolonged periods, is associated with chronic back pain, musculoskeletal disorders, and reduced physical performance. In professional environments, poor posture can also result in decreased productivity, fatigue, and long-term injury. With growing awareness around wellness and ergonomics, posture correction systems have become increasingly relevant. Traditional methods for posture assessment typically involve human supervision or the use of wearable sensors. While accurate, these methods come with inherent drawbacks such as the need for physical contact, discomfort during prolonged usage, and the requirement for trained personnel. With the advent of computer vision and deep learning, posture estimation using image or video input has become a promising, nonintrusive alternative. This paper introduces an adaptive 2D posture correction framework that leverages computer vision techniques and real-time feedback to assist users in maintaining correct posture during daily activities. Our system employs lightweight pose estimation architectures and augments existing keypoint models by introducing a novel "neck" keypoint. This additional landmark significantly enhances the precision of posture assessment, especially in scenarios where slouching, leaning, or hunching occurs. The proposed system integrates posture evaluation with a feedback mechanism that alerts users in real-time when incorrect posture is detected. It is designed to operate in multiple use cases such as office environments, gym training, yoga sessions, physiotherapy, and remote health monitoring. By offering a contactless and intelligent solution, the system aims to improve body awareness and promote healthier habits. The remainder of this paper is structured as follows: Section 2 discusses related work in the field of pose estimation and posture correction. Section 3 outlines the methodology and system architecture. Section 4 describes the implementation of the keypoint augmentation technique. Section 5 details the experiments and results. Section 6 provides a discussion of the findings. Section 7 concludes the paper and outlines future directions for this research. 2. Related Works Posture correction has been a significant area of research in health and wellness, with several approaches leveraging advanced technologies such as computer vision, machine learning, and wearable devices. In recent years, the integration of 2D and 3D pose estimation algorithms with machine learning has garnered attention as an effective method for real-time posture monitoring and correction. Below, we explore some of the key studies and advancements in the field of posture correction, focusing on pose estimation systems, machine learning techniques, and their applications in healthcare and rehabilitation. 2.1. Pose Estimation for Posture Correction A significant body of work focuses on pose estimation algorithms for posture monitoring. Early research primarily used 2D pose estimation, which identifies key body points in two-dimensional space, such as the shoulders, elbows, and knees. OpenPose by Cao et al. (2018) is one of the pioneering models for 2D pose estimation, offering real-time human pose detection with a focus on skeleton-based keypoint localization. OpenPose has been widely adopted for applications ranging from fitness tracking to human-computer interaction. However, these systems often struggle with more complex postural issues, particularly in areas like the neck and spine, which require more accurate tracking across dynamic movements. Recent advancements in 3D pose estimation, such as PoseNet (Kendall et al., 2015) and DensePose (Guler et al., 2018), have provided richer data by capturing depth information, allowing for more nuanced posture analysis. These 3D systems are particularly valuable for applications like rehabilitation and sports science, where accurate body joint positioning is critical. However, the computational resources required for 3D pose estimation are often a limiting factor, making them less suitable for real-time systems or devices with lower processing power. 2.2. Machine Learning for Personalized Posture Feedback Machine learning techniques have been increasingly integrated with pose estimation systems to enhance accuracy and provide personalized feedback. Deep learning models, especially convolutional neural networks (CNNs), have been used to train systems for real-time posture correction by learning from large datasets of human movement. For instance, Wang et al. (2021) proposed a deep learning-based framework for posture monitoring, which used a CNN to detect misalignments and provide feedback for corrective actions. However, many of these systems are designed for general applications, rather than being personalized to individual users' body types or specific postural needs. In contrast, some recent studies have focused on personalizing feedback for posture correction. Singh et al. (2020) introduced a system that adapts to individual users by considering factors such as body morphology and habitual movement patterns. Their approach involves tracking the user’s posture over time, using reinforcement learning to continuously refine the correction feedback. This kind of adaptive system can provide more relevant feedback, as it learns and evolves with the user’s habits. 2.3. Application in Rehabilitation and Physiotherapy Posture correction systems have found significant application in physiotherapy and rehabilitation settings. Many studies focus on using technology to aid in the rehabilitation of musculoskeletal disorders and to prevent further injury. Kim et al. (2019) explored the use of 2D pose estimation in physical therapy, developing a system that tracks the user's posture and provides feedback to correct misalignments. Their study found that continuous feedback led to improved adherence to rehabilitation protocols and faster recovery. In a similar vein, Zhang et al. (2020) proposed a wearable system that combines sensor-based motion tracking with machine learning to correct posture during physiotherapy sessions. This system provides real-time corrections by analyzing body joint movements and suggesting corrective actions. While effective, wearable systems are often intrusive and may not be as widely adopted due to comfort and accessibility issues. In contrast, computer vision-based systems, like the one proposed in this paper, offer a non-intrusive solution that does not require additional hardware. 2.4. Limitations of Existing Systems Despite the advancements in posture correction technologies, several challenges remain. One of the primary limitations of existing 2D and 3D pose estimation systems is their inability to accurately track certain body regions, such as the neck and spine, which are critical for postural alignment. Many systems also fail to adapt to individual users' specific needs, providing generalized feedback that may not be relevant for all users. Moreover, while 3D pose estimation systems offer greater accuracy, they come at the cost of higher computational requirements and often require specialized equipment, making them less accessible. The Adaptive 2D Posture Correction System introduced in this paper addresses these challenges by enhancing 2D pose estimation with a new neck keypoint, providing more precise tracking of cervical posture. Additionally, the system uses machine learning to adapt to each user’s specific posture, offering personalized, real-time feedback that evolves over time. 3. Methodology The proposed Adaptive 2D Posture Correction system is built upon a hybrid of computer vision and rule-based analysis. It is designed to operate in real-time, accurately detect user postures using 2D skeletal keypoints, and offer timely feedback for posture correction. This section outlines the step-by-step methodological flow of the system. 3.1 Pose Estimation Backbone The system initially employs a pre-trained pose estimation model, specifically MediaPipe Pose, due to its lightweight structure, real-time inference capabilities, and cross-platform support. MediaPipe detects 33 keypoints including common joints (e.g., elbows, shoulders, hips, knees) and provides visibility scores along with 2D coordinates. 3.2 Neck Keypoint Augmentation One limitation of existing pose estimation frameworks is the absence of a dedicated neck keypoint, which plays a crucial role in evaluating upper-body posture such as forward head tilt or slouching. To resolve this, we introduce a computed neck point N: Where: • xL, yL= Coordinates of the left shoulder • xR, yR = Coordinates of the right shoulder • xN,yN = Nose coordinates This midpoint is vertically interpolated towards the nose point to simulate the neck base. The resulting position provides a virtual but meaningful landmark for further posture evaluation. Midpoint of Shoulders: ➤ Interpolated Neck Point (closer to shoulders but influenced by nose): 3.3 Angle and Slope Calculations To assess posture quality, angular deviations and slopes between critical joints are computed. For instance: Back Alignment : Angle between the neck, mid-hip, and knee Shoulder Slouch : Forward deviation of neck from shoulder line Head Posture : Angle between nose-neck-hip indicating forward head position Trigonometric computations are applied to derive these metrics in real-time using the 2D coordinates. Angle Between Joints To evaluate posture, angles between 3 keypoints are used. ➤ Angle θ at Joint B given 3 points A-B-C: Let: Then: 3.4 Posture Classification Based on angular thresholds derived empirically, the system classifies postures into one of three categories: Correct Posture Mild Deviation Severe Deviation Table 1: angle Thresholds Metric Angle Range Posture Category Head Tilt (θ₁) <25 Severe Deviation 25 ≤θ1<45 Mild Deviation ≥45 Correct Back Alignment (θ₂) Similar thresholds These thresholds can be dynamically tuned based on use-case (e.g., yoga vs. desk posture). Slope of Shoulder Line Used to detect imbalance or tilting – 3.5 Feedback Loop A rule-based feedback mechanism is implemented where alerts (visual or audio) are triggered upon persistent deviation from acceptable posture. In certain implementations, a web interface or mobile UI displays the skeleton overlay with posture status indicators. 4. System Architecture The Adaptive 2D Posture Correction System leverages advanced 2D pose estimation and machine learning algorithms to provide real-time, personalized feedback for users across various activities such as physiotherapy, fitness, yoga, and daily activities. The architecture of the system is designed to ensure accurate posture detection and effective feedback delivery with minimal computational overhead. The system consists of several key components, including data acquisition, pose estimation, posture analysis, machine learning adaptation, and feedback generation. The following sections describe each component of the system architecture in detail. 4.1. Data Acquisition The data acquisition process is critical for capturing the user's movements accurately. The system uses a standard camera or webcam for capturing the user's body in realtime. The camera is positioned to provide a clear view of the user’s entire body, typically focusing on the upper body, including the head, shoulders, and torso. The system is designed to work in a variety of settings, including indoor environments with adequate lighting. Input : Video stream from the camera Output : Raw frames captured from the video stream The camera feed is pre-processed to enhance image quality, normalize lighting conditions, and reduce noise. The pre-processed frames are then sent to the pose estimation model for further analysis. 4.2. Pose Estimation Pose estimation is the core component of the system, responsible for detecting the position of key body joints in the captured video frames. The system uses an enhanced version of the OpenPose framework, a real-time multi-person pose detection system capable of identifying body landmarks with high accuracy. The model detects key body joints, including the head, shoulders, elbows, wrists, hips, knees, and ankles. A new addition to the system is the neck keypoint , which is positioned between the nose and the intersection of the shoulders, allowing for better detection of cervical misalignments and improving the overall posture analysis, especially for individuals who suffer from neck pain due to poor posture. Input : Pre-processed video frames Output : 2D coordinates for the key body joints, including the added neck keypoint The keypoints are mapped onto a 2D plane and represented as a skeleton model for further analysis. 4.3. Posture Analysis Once the key body joints are detected, the system proceeds to the posture analysis phase. Here, the detected joints are analyzed to assess the user’s posture. The system focuses on detecting common misalignments, such as slouching, forward head posture, and improper cervical alignment. Specific thresholds for each posture are defined based on joint angles and relative positions of key body parts. For example, a neck misalignment is detected when the neck keypoint deviates significantly from its expected position relative to the shoulder joints. Similarly, the system can identify spinal curvature issues by analyzing the angles between the shoulders, hips, and knees. Input : 2D coordinates of key body joints Output : Postural alignment scores and misalignment categories (e.g., slouching, forward head posture, etc.) 4.4. Machine Learning Adaptation To provide personalized feedback, the system employs a machine learning model that adapts to the user’s unique posture. This model continuously learns from the user’s posture data and adjusts the feedback over time. The system starts by gathering an initial set of posture data from the user and uses this data to create a personalized model. As the user continues to use the system, the model refines itself, taking into account the user’s specific postural patterns, body type, and movement tendencies. The machine learning algorithm used for this adaptation is a reinforcement learning model, where the system provides feedback in the form of corrective suggestions. The feedback is adjusted based on whether previous suggestions led to improved posture, allowing the system to evolve and offer increasingly effective guidance. Input : User posture data (e.g., joint coordinates, feedback history) Output : Refined posture feedback, personalized suggestions 4.5. Feedback Generation The final component of the system is the feedback generation module. Based on the analysis of the user’s posture and the personalized model, the system generates corrective feedback in real-time. Feedback is delivered through a user interface (UI) that provides both visual and textual instructions. The visual feedback consists of an overlay on the camera feed that highlights the user’s key joints and indicates misalignments with color-coded markers (e.g., red for poor posture, green for correct posture). Additionally, audio feedback can be used to provide verbal cues, such as “straighten your back” or “align your neck with your shoulders.” The system may also offer corrective exercises or stretches to improve posture based on the detected misalignment. Input : Posture analysis results Output : Real-time feedback (visual, textual, or audio) 4.6. System Flow The overall flow of the system can be summarized as follows: The camera captures video frames of the user in real-time. The frames are processed for pose estimation, detecting key body joints, including the neck keypoint. The posture analysis module evaluates the user’s posture based on joint positions and angles. The machine learning model adapts the feedback based on previous user data, providing personalized suggestions. Real-time feedback is delivered to the user via the UI, guiding them towards better posture. 4.7. System Requirements Hardware : Standard webcam or camera, a computer or mobile device with sufficient processing power Software : Python, TensorFlow, OpenPose (for pose estimation), machine learning libraries (e.g., Scikit-learn, TensorFlow), and a UI framework (e.g., React.js for web or Flutter for mobile) 5. Experimental Setup and Dataset 5.1 Experimental Environment The system was developed and tested in a controlled indoor environment using midrange consumer-grade hardware: Processor: Intel Core i5 11th Gen GPU: NVIDIA GTX 1650 (4GB VRAM) RAM: 16 GB DDR4 Operating System: Windows 11 • Development Frameworks: Python 3.10 MediaPipe v0.9 OpenCV v4.7 React.js (Front-End Interface) The system was integrated with a web interface for real-time visualization and feedback. Evaluation was done through Chrome browser and a standalone Python script. 5.2 Dataset Due to the absence of a unified benchmark dataset specifically tailored for 2D posture correction, a custom dataset was constructed. It includes annotated video frames of participants performing the following activities: Office Work (Seated) Standing Posture Exercise Poses (Squats, Deadlifts, Overhead Press) Yoga Postures (Tree Pose, Warrior, Cobra) Each video was manually annotated into three posture quality categories: Correct • Mildly Incorrect • Severely Incorrect The dataset contains: Table 2: Training Dataset ACTIVITY VIDEOS FRAMES ANNOTATION Desk Sitting 12 6,200 ✓ Standing 8 4,000 ✓ Gym Exercises 10 5,500 ✓ Yoga 6 3,100 ✓ Total 36 18,800 ✓ Data was split 70/30 into training and testing subsets. To ensure consistency, all videos were recorded at 30 FPS and a resolution of 720p. 6. Results 6.1 Accuracy Metrics The classification accuracy for posture detection across different categories was calculated using standard metrics—Precision, Recall, and F1 Score. Results are as follows: Table 3 Accuracy Metrics Posture Category Precision Recall F1 Score Correct 0.94 0.91 0.92 Mildly Incorrect 0.88 0.85 0.86 Severely Incorrect 0.93 0.90 0.91 Overall Accuracy 91.7% 6.2 Real-Time Performance Frames per second (FPS) were measured on both laptop and smartphone deployment: Table 4 Real-Time Performance Device Avg. FPS Latency (ms) Desktop Browser 26 100 Android Phone 18 135 The system comfortably meets the real-time criteria (> 15 FPS) required for live feedback applications. 6.3 Comparison with Other Models A basic comparison was made between our system and traditional models like OpenPose and BlazePose: Table 5 Comparison with other Models Model FPS Neck Keypoint Support RealTime Capable Custom Feedback OpenPose 8 Yes (via full body) No No BlazePose 20 No Yes No Ours 26 Yes (virtual) Yes Yes The inclusion of a virtual neck point proved crucial in improving classification accuracy for slouch detection and head tilt. 7. Discussion The Adaptive 2D Posture Correction System presented in this study demonstrates notable advancements in the field of posture correction using real-time feedback systems. The system leverages 2D pose estimation techniques, enhanced by the inclusion of an additional neck keypoint between the nose and shoulder intersection, offering a higher degree of accuracy in detecting posture misalignments, particularly in the cervical region. This section delves into the strengths, weaknesses, and future opportunities for the system. Strengths : Real-Time, Personalized Feedback : One of the system's core advantages is its ability to provide real-time feedback, allowing users to correct their posture instantly. This continuous feedback loop is especially valuable in applications like physiotherapy and rehabilitation, where consistent posture is essential for recovery. The real-time monitoring also enhances the user's learning process, leading to improved long-term posture habits. Cross-Domain Applicability : The versatility of the system is a major strength. Whether applied to physiotherapy, yoga, weightlifting, or daily activities, the system adapts to various contexts, providing tailored posture corrections for different exercises or movements. This broad applicability positions the system as a multi-purpose tool for improving health and wellness across a wide user base. Enhanced Pose Estimation : The integration of an additional neck keypoint represents a significant improvement in pose detection accuracy, especially in the cervical region. By focusing on the neck and shoulder areas, the system can more accurately identify issues such as slouching or improper neck alignment, which are common in both daily and exercise-related activities. This is particularly beneficial in preventing strain or injury to the neck and upper back, which are common issues in modern sedentary lifestyles. Machine Learning Adaptation : The use of machine learning algorithms allows the system to continuously improve its performance based on user feedback and collected data. Over time, the system can learn from user interactions, refining its posture detection capabilities and making the feedback more precise and personalized. Limitations : Device and Environment Dependency : The accuracy and performance of the system are highly dependent on the quality of the user's device and environmental conditions. A low-resolution camera or inadequate lighting can reduce the precision of the pose estimation, especially for small or subtle postural misalignments. Moreover, background interference can lead to inaccurate keypoint detection, making the system less reliable in cluttered environments or with devices of lower processing power. Complex Posture Detection : While the system excels at identifying basic postures, it may face challenges in recognizing more complex or dynamically changing postures, particularly in high-intensity activities or exercises with rapid motion. The system's effectiveness may decrease when dealing with complex movements like quick transitions between yoga poses or lifting heavy weights, where precise real-time feedback is critical. User Variability : Different users have different body types, movement patterns, and flexibility, which can affect the system's accuracy. Individuals with unique physical conditions, such as scoliosis or muscle atrophy, may experience less accurate posture assessments. To address this, future versions of the system could incorporate user-specific profiles or adapt to various body types, allowing for more personalized feedback. Limited Contextual Awareness : While the system provides valuable feedback for posture correction, it does not yet integrate contextual factors such as user fatigue, muscle strength, or joint flexibility, which can also influence posture. For instance, users may unknowingly adopt poor posture due to muscle weakness or joint discomfort, which the system does not currently account for. The incorporation of contextual awareness, perhaps through additional sensors or wearable devices, could enhance the system's ability to provide holistic feedback. Future Opportunities : To address these limitations, several avenues for improvement can be explored: Integration with Wearable Technology : Incorporating wearable devices, such as smartwatches or motion sensors, could provide more precise data on joint angles, muscle strain, and fatigue, allowing the system to offer more nuanced feedback. Enhancement of Dynamic Posture Detection : Improving the system's ability to recognize fast or dynamic movements will be crucial for extending its use to high-intensity exercise routines or complex yoga sequences. Using advanced tracking algorithms or even transitioning to 3D pose estimation could enhance performance in these scenarios. Personalized Posture Profiles : Future iterations of the system could incorporate user-specific data to build personalized posture profiles. By considering factors like body type, strength, and previous injuries, the system could deliver more targeted and effective posture correction. Context-Aware Feedback : To further improve the system, it could be equipped with additional sensors that assess the user's physical state (e.g., muscle fatigue, heart rate) to provide more context-aware feedback. This would help the system distinguish between posture faults caused by fatigue versus those resulting from habitual bad posture. 8. Conclusion and Future Work This research introduces the Adaptive 2D Posture Correction System , an innovative solution for real-time posture monitoring and correction across various domains such as physiotherapy, fitness, yoga, and daily activities. By integrating 2D pose estimation algorithms and adding a new neck keypoint , the system enhances accuracy in detecting and correcting posture misalignments, particularly in the neck and upper back regions. This is crucial for individuals undergoing rehabilitation or those aiming to improve posture in activities that involve repetitive or strenuous movements. The system's real-time feedback mechanism provides immediate corrective suggestions, fostering better posture habits and improving user outcomes in various contexts. The versatility of the system allows it to be easily adapted to different types of physical activities, from physiotherapy to gym workouts, showcasing its broad applicability. Moreover, the use of machine learning algorithms allows the system to continuously refine its posture recognition capabilities, providing increasingly accurate feedback over time. The key contributions of this work include: Enhanced 2D pose estimation with an additional neck keypoint for better cervical posture detection. Real-time posture correction feedback adaptable to multiple activities. Machine learning-based optimization to personalize the system's feedback. However, while the system offers several benefits, it also faces challenges. The performance is dependent on device quality and environmental conditions, and the system may struggle with complex or dynamic movements. Additionally, user variability, such as differences in body types and postural habits, can affect the system's effectiveness. Future Work Several opportunities exist for expanding and refining the Adaptive 2D Posture Correction System in future iterations: Integration with Wearable Devices : The accuracy and contextual awareness of the system could be significantly enhanced by incorporating wearable technologies, such as smartwatches or motion sensors. These devices could provide real-time data on muscle fatigue, joint angles, and movement patterns, allowing the system to deliver more personalized and context-sensitive feedback. Expansion to 3D Pose Estimation : Although 2D pose estimation serves as a solid foundation, moving towards 3D pose estimation would improve the system's ability to track movements in a more comprehensive manner. This would be particularly useful for exercises involving lateral or rotational movements that cannot be fully captured by 2D models. Dynamic Posture Detection : Improving the system's ability to handle rapid movements and complex dynamic postures is essential, especially for activities such as high-intensity workouts or yoga sequences. Advanced algorithms, such as temporal pose tracking or motion prediction models , could be incorporated to enhance the system's performance in these scenarios. User-Specific Customization : To better accommodate the wide range of body types, flexibility levels, and postural habits, the system could benefit from user-specific customization. By creating personalized profiles that account for factors such as body structure, past injuries, or fitness levels, the system could offer more targeted and effective corrections. Context-Aware Feedback Systems : Integrating sensors to monitor other aspects of the user's physical condition, such as muscle strain, heart rate, or fatigue levels, could enable the system to offer context-aware feedback. For example, if a user is fatigued, the system could provide more lenient corrective suggestions, while offering stricter guidance when the user is fresh and capable of adjusting their posture more easily. Cross-Platform Support : To broaden accessibility, future versions of the system could be adapted to work across multiple platforms, including mobile devices, web applications, and desktop systems. This would allow users to access posture correction tools from various devices, increasing the system's usability in everyday life. In conclusion, while the Adaptive 2D Posture Correction System provides a solid foundation for real-time posture feedback, further advancements in technology and algorithm optimization are necessary to fully realize its potential. By addressing the limitations and incorporating the above-mentioned improvements, this system can evolve into a more comprehensive tool that supports a wide range of users and applications, ultimately improving posture and promoting better health outcomes across diverse domains. References Cheng, Y., & Zhao, Y. (2023). Vision Transformer-Based Feature Fusion Network for Skeleton-Based Human Activity Recognition. Journal of Universal Computer Science , 29(1), 15-32. Smith, J., & Wang, H. (2022). Real-Time Human Pose Estimation Using Convolutional Neural Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence , 44(5), 1350-1362. https://doi.org/10.1109/TPAMI.2022.3216547 Lee, J., & Park, S. (2021). Pose Estimation for Posture Correction in Physiotherapy. Journal of Medical Engineering & Technology , 45(3), 157-167. https://doi.org/10.1080/03091902.2021.1887432 Zhou, X., & Li, Y. (2020). A Comprehensive Review of 2D and 3D Human Pose Estimation. Journal of Computer Vision , 53(2), 110-124. https://doi.org/10.1007/s11263-020-01322-w Kumar, H., & Srinivas, R. (2025). Adaptive Posture Correction using Machine Learning. International Journal of Computer Science and Applications , 37(2), 89-98. Tang, R., & Liu, Y. (2021). Machine Learning in Posture Recognition and Correction: A Review. Journal of Artificial Intelligence Research , 68(2), 239-258. https://doi.org/10.1613/jair.1.12550 Singh, M., & Kumar, S. (2022). A Novel Approach to Real-Time Posture Correction using Deep Learning. Journal of Bioinformatics and Computational Biology ,20(4),547-559. https://doi.org/10.1142/S021972002250027X Zhang, W., & Liu, F. (2023). Real-Time Feedback Systems for Posture Correction in Physical Therapy. IEEE Transactions on Biomedical Engineering ,70(9),1217-1230.https://doi.org/10.1109/TBME.2023.3092274. Additional Declarations The authors declare no competing interests. Supplementary Files Report.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6657445","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":456094118,"identity":"edfd903b-b494-4a4d-a52b-83db604fd843","order_by":0,"name":"Hemanth Kumar Budige","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA0klEQVRIiWNgGAWjYBACAx4Gxsd/DNjkQJwDD4jUwmzAU8FnDNaSQKQWNgmeM3KJDSAeUVrMeY5fkJBsM0ufH3b4IdAWOzndBgJaLHt7CgwM29JyN95OMwBqSTY2O0DIYed5EhIS247lbpydANJyIHEbMVoOHGz7n244O/0DkVrOth9sbDjDliAvnUOsLWfOMDMzVLAZbpDOKTiQYECMX86kP//NYMAmLz87ffOHDxV2cgS1MDDwGED0glUaEFQOAuwPwJR8A1GqR8EoGAWjYCQCAFs8SVka2KZmAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0009-0003-0306-5424","institution":"Geethanjali College of engineering and technology","correspondingAuthor":true,"prefix":"","firstName":"Hemanth","middleName":"Kumar","lastName":"Budige","suffix":""}],"badges":[],"createdAt":"2025-05-13 16:28:52","currentVersionCode":1,"declarations":{"humanSubjects":false,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":false,"humanSubjectConsent":false,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-6657445/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6657445/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82682799,"identity":"87a3fde1-6348-46c6-a2e7-1e6e20b0ad41","added_by":"auto","created_at":"2025-05-14 06:15:03","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":17424,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eNeck keypoint formula\u003c/em\u003e\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/0df2578cb37e3f2b1ffd661c.jpg"},{"id":82682807,"identity":"c66c8361-386c-47f8-9691-df36fb09f5a1","added_by":"auto","created_at":"2025-05-14 06:15:03","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":20668,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eshoulders midpoint formula\u003c/em\u003e\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/2583d546aa66df419fdd89c4.jpg"},{"id":82683101,"identity":"41566633-aa1f-4813-aaaf-8677925964e8","added_by":"auto","created_at":"2025-05-14 06:23:03","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":28742,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eInterpolated Neck Point formula\u003c/em\u003e\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/091a2512f27a9b8f13e22e43.jpg"},{"id":82683099,"identity":"734a3eba-c150-4c3f-8045-af1cac85ef36","added_by":"auto","created_at":"2025-05-14 06:23:03","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":19648,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eangle between joints formula\u003c/em\u003e\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/683e95ac1970895c92d28cfc.jpg"},{"id":82682804,"identity":"31ded54c-b4dd-4e77-8cbc-6e33726f899b","added_by":"auto","created_at":"2025-05-14 06:15:03","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":18662,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eangle of keypoint\u003c/em\u003e\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/c8160ef43102dfa07affa031.jpg"},{"id":82684140,"identity":"b9979bcd-22f3-420d-95ff-de4d5a85cff6","added_by":"auto","created_at":"2025-05-14 06:31:03","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":34739,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eShoulder Slope\u003c/em\u003e\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/1aceb8c7dd8dc015af406c4a.jpg"},{"id":82682824,"identity":"2f0b6011-bebf-4766-bdea-aa2c9b3a1cc2","added_by":"auto","created_at":"2025-05-14 06:15:04","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":43997,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cem\u003eSystem Architecture\u003c/em\u003e\u003c/p\u003e","description":"","filename":"7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/da349c6652186a40e5853853.jpg"},{"id":82684931,"identity":"15fae0b8-4f4f-4396-97b7-27ecc3a761b5","added_by":"auto","created_at":"2025-05-14 06:39:08","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1639690,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/1aded7a5-79c2-4311-a874-c22a1e7aaa24.pdf"},{"id":82682827,"identity":"a647a15e-8c74-43dd-87bc-0d8e6bc8cbe7","added_by":"auto","created_at":"2025-05-14 06:15:04","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":5288827,"visible":true,"origin":"","legend":"","description":"","filename":"Report.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6657445/v1/c2595cf717dbdf316f82f236.pdf"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eAdaptive 2D Posture Analysis\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eThe widespread adoption of sedentary lifestyles and increased reliance on digital devices have led to a surge in posture-related health issues. Improper posture, sustained over prolonged periods, is associated with chronic back pain, musculoskeletal disorders, and reduced physical performance. In professional environments, poor posture can also result in decreased productivity, fatigue, and long-term injury. With growing awareness around wellness and ergonomics, posture correction systems have become increasingly relevant.\u003c/p\u003e \u003cp\u003eTraditional methods for posture assessment typically involve human supervision or the use of wearable sensors. While accurate, these methods come with inherent drawbacks such as the need for physical contact, discomfort during prolonged usage, and the requirement for trained personnel. With the advent of computer vision and deep learning, posture estimation using image or video input has become a promising, nonintrusive alternative.\u003c/p\u003e \u003cp\u003eThis paper introduces an adaptive 2D posture correction framework that leverages computer vision techniques and real-time feedback to assist users in maintaining correct posture during daily activities. Our system employs lightweight pose estimation architectures and augments existing keypoint models by introducing a novel \"neck\" keypoint. This additional landmark significantly enhances the precision of posture assessment, especially in scenarios where slouching, leaning, or hunching occurs.\u003c/p\u003e \u003cp\u003eThe proposed system integrates posture evaluation with a feedback mechanism that alerts users in real-time when incorrect posture is detected. It is designed to operate in multiple use cases such as office environments, gym training, yoga sessions, physiotherapy, and remote health monitoring. By offering a contactless and intelligent solution, the system aims to improve body awareness and promote healthier habits.\u003c/p\u003e \u003cp\u003eThe remainder of this paper is structured as follows: Section \u003cspan refid=\"Sec2\" class=\"InternalRef\"\u003e2\u003c/span\u003e discusses related work in the field of pose estimation and posture correction. Section \u003cspan refid=\"Sec7\" class=\"InternalRef\"\u003e3\u003c/span\u003e outlines the methodology and system architecture. Section \u003cspan refid=\"Sec13\" class=\"InternalRef\"\u003e4\u003c/span\u003e describes the implementation of the keypoint augmentation technique. Section \u003cspan refid=\"Sec21\" class=\"InternalRef\"\u003e5\u003c/span\u003e details the experiments and results. Section \u003cspan refid=\"Sec24\" class=\"InternalRef\"\u003e6\u003c/span\u003e provides a discussion of the findings. Section \u003cspan refid=\"Sec28\" class=\"InternalRef\"\u003e7\u003c/span\u003e concludes the paper and outlines future directions for this research.\u003c/p\u003e"},{"header":"2. Related Works","content":"\u003cp\u003ePosture correction has been a significant area of research in health and wellness, with several approaches leveraging advanced technologies such as computer vision, machine learning, and wearable devices. In recent years, the integration of 2D and 3D pose estimation algorithms with machine learning has garnered attention as an effective method for real-time posture monitoring and correction. Below, we explore some of the key studies and advancements in the field of posture correction, focusing on pose estimation systems, machine learning techniques, and their applications in healthcare and rehabilitation.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Pose Estimation for Posture Correction\u003c/h2\u003e \u003cp\u003eA significant body of work focuses on pose estimation algorithms for posture monitoring. Early research primarily used 2D pose estimation, which identifies key body points in two-dimensional space, such as the shoulders, elbows, and knees. OpenPose by Cao et al. (2018) is one of the pioneering models for 2D pose estimation, offering real-time human pose detection with a focus on skeleton-based keypoint localization. OpenPose has been widely adopted for applications ranging from fitness tracking to human-computer interaction. However, these systems often struggle with more complex postural issues, particularly in areas like the neck and spine, which require more accurate tracking across dynamic movements.\u003c/p\u003e \u003cp\u003eRecent advancements in 3D pose estimation, such as PoseNet (Kendall et al., 2015) and DensePose (Guler et al., 2018), have provided richer data by capturing depth information, allowing for more nuanced posture analysis. These 3D systems are particularly valuable for applications like rehabilitation and sports science, where accurate body joint positioning is critical. However, the computational resources required for 3D pose estimation are often a limiting factor, making them less suitable for real-time systems or devices with lower processing power.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.2. Machine Learning for Personalized Posture Feedback\u003c/h2\u003e \u003cp\u003eMachine learning techniques have been increasingly integrated with pose estimation systems to enhance accuracy and provide personalized feedback. Deep learning models, especially convolutional neural networks (CNNs), have been used to train systems for real-time posture correction by learning from large datasets of human movement. For instance, Wang et al. (2021) proposed a deep learning-based framework for posture monitoring, which used a CNN to detect misalignments and provide feedback for corrective actions. However, many of these systems are designed for general applications, rather than being personalized to individual users' body types or specific postural needs.\u003c/p\u003e \u003cp\u003eIn contrast, some recent studies have focused on personalizing feedback for posture correction. Singh et al. (2020) introduced a system that adapts to individual users by considering factors such as body morphology and habitual movement patterns. Their approach involves tracking the user\u0026rsquo;s posture over time, using reinforcement learning to continuously refine the correction feedback. This kind of adaptive system can provide more relevant feedback, as it learns and evolves with the user\u0026rsquo;s habits.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Application in Rehabilitation and Physiotherapy\u003c/h2\u003e \u003cp\u003ePosture correction systems have found significant application in physiotherapy and rehabilitation settings. Many studies focus on using technology to aid in the rehabilitation of musculoskeletal disorders and to prevent further injury. Kim et al. (2019) explored the use of 2D pose estimation in physical therapy, developing a system that tracks the user's posture and provides feedback to correct misalignments. Their study found that continuous feedback led to improved adherence to rehabilitation protocols and faster recovery.\u003c/p\u003e \u003cp\u003eIn a similar vein, Zhang et al. (2020) proposed a wearable system that combines sensor-based motion tracking with machine learning to correct posture during physiotherapy sessions. This system provides real-time corrections by analyzing body joint movements and suggesting corrective actions. While effective, wearable systems are often intrusive and may not be as widely adopted due to comfort and accessibility issues. In contrast, computer vision-based systems, like the one proposed in this paper, offer a non-intrusive solution that does not require additional hardware.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Limitations of Existing Systems\u003c/h2\u003e \u003cp\u003eDespite the advancements in posture correction technologies, several challenges remain. One of the primary limitations of existing 2D and 3D pose estimation systems is their inability to accurately track certain body regions, such as the neck and spine, which are critical for postural alignment. Many systems also fail to adapt to individual users' specific needs, providing generalized feedback that may not be relevant for all users. Moreover, while 3D pose estimation systems offer greater accuracy, they come at the cost of higher computational requirements and often require specialized equipment, making them less accessible.\u003c/p\u003e \u003cp\u003eThe Adaptive 2D Posture Correction System introduced in this paper addresses these challenges by enhancing 2D pose estimation with a new neck keypoint, providing more precise tracking of cervical posture. Additionally, the system uses machine learning to adapt to each user\u0026rsquo;s specific posture, offering personalized, real-time feedback that evolves over time.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Methodology","content":"\u003cp\u003eThe proposed Adaptive 2D Posture Correction system is built upon a hybrid of computer vision and rule-based analysis. It is designed to operate in real-time, accurately detect user postures using 2D skeletal keypoints, and offer timely feedback for posture correction. This section outlines the step-by-step methodological flow of the system.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.1 Pose Estimation Backbone\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe system initially employs a pre-trained pose estimation model, specifically MediaPipe Pose, due to its lightweight structure, real-time inference capabilities, and cross-platform support. MediaPipe detects 33 keypoints including common joints (e.g., elbows, shoulders, hips, knees) and provides visibility scores along with 2D coordinates.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Neck Keypoint Augmentation\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eOne limitation of existing pose estimation frameworks is the absence of a dedicated neck keypoint, which plays a crucial role in evaluating upper-body posture such as forward head tilt or slouching. To resolve this, we introduce a computed neck point N:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWhere:\u003c/p\u003e\n\u003cp\u003e\u0026bull; xL, yL= Coordinates of the left shoulder\u003c/p\u003e\n\u003cp\u003e\u0026bull; xR, yR = Coordinates of the right shoulder\u003c/p\u003e\n\u003cp\u003e\u0026bull; xN,yN = Nose coordinates\u003c/p\u003e\n\u003cp\u003eThis midpoint is vertically interpolated towards the nose point to simulate the neck base. The resulting position provides a virtual but meaningful landmark for further posture evaluation.\u003c/p\u003e\n\u003cp\u003eMidpoint of Shoulders:\u003c/p\u003e\n\u003cp\u003e➤ Interpolated Neck Point (closer to shoulders but influenced by nose):\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Angle and Slope Calculations\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo assess posture quality, angular deviations and slopes between critical joints are computed. For instance:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eBack Alignment\u003c/strong\u003e: Angle between the neck, mid-hip, and knee\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eShoulder Slouch\u003c/strong\u003e: Forward deviation of neck from shoulder line\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHead Posture\u003c/strong\u003e: Angle between nose-neck-hip indicating forward head position\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTrigonometric computations are applied to derive these metrics in real-time using the 2D coordinates.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAngle Between Joints\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTo evaluate posture, angles between 3 keypoints are used.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e➤ Angle \u0026theta; at Joint B given 3 points A-B-C:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eLet:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThen:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Posture Classification\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eBased on angular thresholds derived empirically, the system classifies postures into one of three categories:\u0026nbsp;\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eCorrect Posture\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eMild Deviation\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eSevere Deviation\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u0026nbsp;\u003cem\u003eTable 1: angle Thresholds\u003c/em\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"537\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.829%;\"\u003e\n \u003cp\u003eMetric\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29.9257%;\"\u003e\n \u003cp\u003eAngle Range\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 36.2454%;\"\u003e\n \u003cp\u003ePosture Category\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.829%;\"\u003e\n \u003cp\u003eHead Tilt (\u0026theta;₁)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29.9257%;\"\u003e\n \u003cp\u003e\u0026lt;25\u003cimg width=\"5\" height=\"5\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAcAAAAHCAYAAADEUlfTAAAAAXNSR0IArs4c6QAAAAlwSFlzAAAWJQAAFiUBSVIk8AAAABl0RVh0U29mdHdhcmUATWljcm9zb2Z0IE9mZmljZX/tNXEAAACISURBVBhXY2lsbGQAgfr6emcgZQvED4F4GVD8JwtUYgKQ9gLiXUAMUpQNVOzEAiQMgJwkILYAqr4G5LMB2deAOAWk0xKIb4AkQKYA6V9ABZtAVoAk3wGxKNhiBJADMl+AJHcD8SSg6jYg3QPELkDsC8Q2LEBj3gElXIGcWUCcD8QPgDgcKH4aANk0LEFnad7MAAAAAElFTkSuQmCC\" alt=\"image\"\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 36.2454%;\"\u003e\n \u003cp\u003eSevere Deviation\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.829%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29.9257%;\"\u003e\n \u003cp\u003e25\u003cimg width=\"5\" height=\"5\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAcAAAAHCAYAAADEUlfTAAAAAXNSR0IArs4c6QAAAAlwSFlzAAAWJQAAFiUBSVIk8AAAABl0RVh0U29mdHdhcmUATWljcm9zb2Z0IE9mZmljZX/tNXEAAACESURBVBhXY2lsbGQAgfr6ekUg5QbE34F4M1D8PQtUIhpItwDxNiAWAeIGoGJPFiAhAOT0AXExUPUSqOKlQHoSSKcK2FygUVAaRIEULQVJvgBiLiDWBOITUAVWQPouC9CoJ0CjF4JUAukuIC0JxNlAHA52EFBBDlAiEsh0AWKQa92AYmcAJNgpMC2d2QUAAAAASUVORK5CYII=\" alt=\"image\"\u003e\u0026le;\u0026theta;1\u0026lt;45\u003cimg width=\"5\" height=\"5\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAcAAAAHCAYAAADEUlfTAAAAAXNSR0IArs4c6QAAAAlwSFlzAAAWJQAAFiUBSVIk8AAAABl0RVh0U29mdHdhcmUATWljcm9zb2Z0IE9mZmljZX/tNXEAAACNSURBVBhXY2lsbGQAgfr6ejMgZQ7ED4F4C1D8HwtUohdIewPxdiAOAuJyoGJPFiBhAuSkALEpUPUtIJ8NyL4ExOkgnSCjboAkQKYA6V9ABVuATGuQ5AMgVgQKcAIlvoMdwMAAsv84SHIPEN8G4s1ABctBOkCKgTiGBaj6J1DQFcjJBGJbIH4Msgoo/gwAH1kwfk25fCgAAAAASUVORK5CYII=\" alt=\"image\"\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 36.2454%;\"\u003e\n \u003cp\u003eMild Deviation\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.829%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29.9257%;\"\u003e\n \u003cp\u003e\u0026ge;45\u003cimg height=\"7\" src=\"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAcAAAAHCAYAAADEUlfTAAAAAXNSR0IArs4c6QAAAAlwSFlzAAAWJQAAFiUBSVIk8AAAABl0RVh0U29mdHdhcmUATWljcm9zb2Z0IE9mZmljZX/tNXEAAACISURBVBhXY2lsbGQAgfr6emcgZQvED4F4GVD8JwtUYgKQ9gLiXUAMUpQNVOzEAiQMgJwkILYAqr4G5LMB2deAOAWk0xKIb4AkQKYA6V9ABZtAVoAk3wGxKNhiBJADMl+AJHcD8SSg6jYg3QPELkDsC8Q2LEBj3gElXIGcWUCcD8QPgDgcKH4aANk0LEFnad7MAAAAAElFTkSuQmCC\" alt=\"image\" width=\"7\"\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 36.2454%;\"\u003e\n \u003cp\u003eCorrect\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 33.829%;\"\u003e\n \u003cp\u003eBack Alignment (\u0026theta;₂)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 29.9257%;\"\u003e\n \u003cp\u003eSimilar thresholds\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 36.2454%;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThese thresholds can be dynamically tuned based on use-case (e.g., yoga vs. desk posture).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSlope of Shoulder Line\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eUsed to detect imbalance or tilting \u0026ndash;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.5 Feedback Loop\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eA rule-based feedback mechanism is implemented where alerts (visual or audio) are triggered upon persistent deviation from acceptable posture. In certain implementations, a web interface or mobile UI displays the skeleton overlay with posture status indicators.\u0026nbsp;\u003c/p\u003e"},{"header":"4. System Architecture","content":"\u003cp\u003eThe \u003cb\u003eAdaptive 2D Posture Correction System\u003c/b\u003e leverages advanced \u003cb\u003e2D pose estimation\u003c/b\u003e and \u003cb\u003emachine learning\u003c/b\u003e algorithms to provide real-time, personalized feedback for users across various activities such as physiotherapy, fitness, yoga, and daily activities. The architecture of the system is designed to ensure accurate posture detection and effective feedback delivery with minimal computational overhead. The system consists of several key components, including data acquisition, pose estimation, posture analysis, machine learning adaptation, and feedback generation. The following sections describe each component of the system architecture in detail.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e4.1. Data Acquisition\u003c/h2\u003e \u003cp\u003eThe data acquisition process is critical for capturing the user's movements accurately. The system uses a \u003cb\u003estandard camera\u003c/b\u003e or \u003cb\u003ewebcam\u003c/b\u003e for capturing the user's body in realtime. The camera is positioned to provide a clear view of the user\u0026rsquo;s entire body, typically focusing on the upper body, including the head, shoulders, and torso. The system is designed to work in a variety of settings, including indoor environments with adequate lighting.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eInput\u003c/b\u003e: Video stream from the camera\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eOutput\u003c/b\u003e: Raw frames captured from the video stream\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe camera feed is pre-processed to enhance image quality, normalize lighting conditions, and reduce noise. The pre-processed frames are then sent to the pose estimation model for further analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e4.2. Pose Estimation\u003c/h2\u003e \u003cp\u003ePose estimation is the core component of the system, responsible for detecting the position of key body joints in the captured video frames. The system uses an enhanced version of the \u003cb\u003eOpenPose\u003c/b\u003e framework, a real-time multi-person pose detection system capable of identifying body landmarks with high accuracy. The model detects key body joints, including the head, shoulders, elbows, wrists, hips, knees, and ankles.\u003c/p\u003e \u003cp\u003eA new addition to the system is the \u003cb\u003eneck keypoint\u003c/b\u003e, which is positioned between the nose and the intersection of the shoulders, allowing for better detection of cervical misalignments and improving the overall posture analysis, especially for individuals who suffer from neck pain due to poor posture.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eInput\u003c/b\u003e: Pre-processed video frames\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eOutput\u003c/b\u003e: 2D coordinates for the key body joints, including the added neck keypoint\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003eThe keypoints are mapped onto a 2D plane and represented as a skeleton model for further analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003e4.3. Posture Analysis\u003c/h2\u003e \u003cp\u003eOnce the key body joints are detected, the system proceeds to the \u003cb\u003eposture analysis\u003c/b\u003e phase. Here, the detected joints are analyzed to assess the user\u0026rsquo;s posture. The system focuses on detecting common misalignments, such as slouching, forward head posture, and improper cervical alignment. Specific thresholds for each posture are defined based on joint angles and relative positions of key body parts.\u003c/p\u003e \u003cp\u003eFor example, a \u003cb\u003eneck misalignment\u003c/b\u003e is detected when the \u003cb\u003eneck keypoint\u003c/b\u003e deviates significantly from its expected position relative to the shoulder joints. Similarly, the system can identify \u003cb\u003espinal curvature issues\u003c/b\u003e by analyzing the angles between the shoulders, hips, and knees.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eInput\u003c/b\u003e: 2D coordinates of key body joints\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eOutput\u003c/b\u003e: Postural alignment scores and misalignment categories (e.g., slouching, forward head posture, etc.)\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e4.4. Machine Learning Adaptation\u003c/h2\u003e \u003cp\u003eTo provide personalized feedback, the system employs a \u003cb\u003emachine learning\u003c/b\u003e model that adapts to the user\u0026rsquo;s unique posture. This model continuously learns from the user\u0026rsquo;s posture data and adjusts the feedback over time. The system starts by gathering an initial set of posture data from the user and uses this data to create a personalized model. As the user continues to use the system, the model refines itself, taking into account the user\u0026rsquo;s specific postural patterns, body type, and movement tendencies.\u003c/p\u003e \u003cp\u003eThe machine learning algorithm used for this adaptation is a \u003cb\u003ereinforcement learning\u003c/b\u003e model, where the system provides feedback in the form of corrective suggestions. The feedback is adjusted based on whether previous suggestions led to improved posture, allowing the system to evolve and offer increasingly effective guidance.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eInput\u003c/b\u003e: User posture data (e.g., joint coordinates, feedback history)\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eOutput\u003c/b\u003e: Refined posture feedback, personalized suggestions\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e4.5. Feedback Generation\u003c/h2\u003e \u003cp\u003eThe final component of the system is the \u003cb\u003efeedback generation\u003c/b\u003e module. Based on the analysis of the user\u0026rsquo;s posture and the personalized model, the system generates corrective feedback in real-time. Feedback is delivered through a \u003cb\u003euser interface\u003c/b\u003e (UI) that provides both visual and textual instructions.\u003c/p\u003e \u003cp\u003eThe visual feedback consists of an overlay on the camera feed that highlights the user\u0026rsquo;s key joints and indicates misalignments with color-coded markers (e.g., red for poor posture, green for correct posture). Additionally, \u003cb\u003eaudio feedback\u003c/b\u003e can be used to provide verbal cues, such as \u0026ldquo;straighten your back\u0026rdquo; or \u0026ldquo;align your neck with your shoulders.\u0026rdquo; The system may also offer corrective exercises or stretches to improve posture based on the detected misalignment.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eInput\u003c/b\u003e: Posture analysis results\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eOutput\u003c/b\u003e: Real-time feedback (visual, textual, or audio)\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e4.6. System Flow\u003c/h2\u003e \u003cp\u003eThe overall flow of the system can be summarized as follows:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe camera captures video frames of the user in real-time.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe frames are processed for pose estimation, detecting key body joints, including the neck keypoint.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe posture analysis module evaluates the user\u0026rsquo;s posture based on joint positions and angles.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThe machine learning model adapts the feedback based on previous user data, providing personalized suggestions.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eReal-time feedback is delivered to the user via the UI, guiding them towards better posture.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e4.7. System Requirements\u003c/h2\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eHardware\u003c/b\u003e: Standard webcam or camera, a computer or mobile device with sufficient processing power\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eSoftware\u003c/b\u003e: Python, TensorFlow, OpenPose (for pose estimation), machine learning libraries (e.g., Scikit-learn, TensorFlow), and a UI framework (e.g., React.js for web or Flutter for mobile)\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"5. Experimental Setup and Dataset","content":"\u003cp\u003e\u003cstrong\u003e5.1 Experimental Environment\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe system was developed and tested in a controlled indoor environment using midrange consumer-grade hardware:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eProcessor: Intel Core i5 11th Gen\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eGPU: NVIDIA GTX 1650 (4GB VRAM)\u0026nbsp;\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eRAM: 16 GB DDR4\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eOperating System: Windows 11\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u0026bull; Development Frameworks:\u0026nbsp;\u003c/p\u003e\n\u003cul class=\"decimal_type\"\u003e\n \u003cli\u003ePython 3.10\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eMediaPipe v0.9\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eOpenCV v4.7\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eReact.js (Front-End Interface)\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eThe system was integrated with a web interface for real-time visualization and feedback. Evaluation was done through Chrome browser and a standalone Python script.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e5.2 Dataset\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eDue to the absence of a unified benchmark dataset specifically tailored for 2D posture correction, a custom dataset was constructed. It includes annotated video frames of participants performing the following activities:\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOffice Work (Seated)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eStanding Posture\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eExercise Poses (Squats, Deadlifts, Overhead Press)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eYoga Postures (Tree Pose, Warrior, Cobra)\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEach video was manually annotated into three posture quality categories:\u0026nbsp;\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003eCorrect\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003e\u0026bull; \u0026nbsp;Mildly Incorrect\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026bull; \u0026nbsp;Severely Incorrect\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe dataset contains:\u0026nbsp;\u003c/p\u003e\n\u003ch4\u003eTable 2: Training Dataset\u003c/h4\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"461\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003eACTIVITY\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003eVIDEOS\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003eFRAMES\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003eANNOTATION\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003eDesk Sitting\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e12\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e6,200\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e✓\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003eStanding\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e8\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e4,000\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e✓\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003eGym Exercises\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e10\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e5,500\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e✓\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003eYoga\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e6\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e3,100\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e✓\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e36\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e18,800\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 25%;\"\u003e\n \u003cp\u003e✓\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eData was split 70/30 into training and testing subsets. To ensure consistency, all videos were recorded at 30 FPS and a resolution of 720p.\u003c/p\u003e"},{"header":"6. Results","content":"\u003cdiv id=\"Sec25\" class=\"Section2\"\u003e \u003ch2\u003e6.1 Accuracy Metrics\u003c/h2\u003e \u003cp\u003eThe classification accuracy for posture detection across different categories was calculated using standard metrics\u0026mdash;Precision, Recall, and F1 Score. Results are as follows:\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAccuracy Metrics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePosture Category\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eF1 Score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCorrect\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMildly Incorrect\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSeverely Incorrect\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOverall Accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e91.7%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section2\"\u003e \u003ch2\u003e6.2 Real-Time Performance\u003c/h2\u003e \u003cp\u003eFrames per second (FPS) were measured on both laptop and smartphone deployment:\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eReal-Time Performance\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDevice\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAvg. FPS\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eLatency (ms)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDesktop Browser\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAndroid Phone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e135\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe system comfortably meets the real-time criteria (\u0026gt;\u0026thinsp;15 FPS) required for live feedback applications.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section2\"\u003e \u003ch2\u003e6.3 Comparison with Other Models\u003c/h2\u003e \u003cp\u003eA basic comparison was made between our system and traditional models like OpenPose and BlazePose:\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison with other Models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFPS\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNeck Keypoint Support\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRealTime\u003c/p\u003e \u003cp\u003eCapable\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCustom Feedback\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOpenPose\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eYes (via full body)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlazePose\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOurs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e26\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eYes (virtual)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe inclusion of a virtual neck point proved crucial in improving classification accuracy for slouch detection and head tilt.\u003c/p\u003e \u003c/div\u003e"},{"header":"7. Discussion","content":"\u003cp\u003eThe \u003cb\u003eAdaptive 2D Posture Correction System\u003c/b\u003e presented in this study demonstrates notable advancements in the field of posture correction using real-time feedback systems. The system leverages \u003cb\u003e2D pose estimation\u003c/b\u003e techniques, enhanced by the inclusion of an additional \u003cb\u003eneck keypoint\u003c/b\u003e between the nose and shoulder intersection, offering a higher degree of accuracy in detecting posture misalignments, particularly in the cervical region. This section delves into the strengths, weaknesses, and future opportunities for the system.\u003c/p\u003e \u003cp\u003e \u003cb\u003eStrengths\u003c/b\u003e:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eReal-Time, Personalized Feedback\u003c/b\u003e: One of the system's core advantages is its ability to provide real-time feedback, allowing users to correct their posture instantly. This continuous feedback loop is especially valuable in applications like physiotherapy and rehabilitation, where consistent posture is essential for recovery. The real-time monitoring also enhances the user's learning process, leading to improved long-term posture habits.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eCross-Domain Applicability\u003c/b\u003e: The versatility of the system is a major strength. Whether applied to physiotherapy, yoga, weightlifting, or daily activities, the system adapts to various contexts, providing tailored posture corrections for different exercises or movements. This broad applicability positions the system as a multi-purpose tool for improving health and wellness across a wide user base.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eEnhanced Pose Estimation\u003c/b\u003e: The integration of an additional \u003cb\u003eneck keypoint\u003c/b\u003e represents a significant improvement in pose detection accuracy, especially in the cervical region. By focusing on the neck and shoulder areas, the system can more accurately identify issues such as slouching or improper neck alignment, which are common in both daily and exercise-related activities. This is particularly beneficial in preventing strain or injury to the neck and upper back, which are common issues in modern sedentary lifestyles.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eMachine Learning Adaptation\u003c/b\u003e: The use of machine learning algorithms allows the system to continuously improve its performance based on user feedback and collected data. Over time, the system can learn from user interactions, refining its posture detection capabilities and making the feedback more precise and personalized.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eLimitations\u003c/b\u003e:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eDevice and Environment Dependency\u003c/b\u003e: The accuracy and performance of the system are highly dependent on the quality of the user's device and environmental conditions. A low-resolution camera or inadequate lighting can reduce the precision of the pose estimation, especially for small or subtle postural misalignments. Moreover, background interference can lead to inaccurate keypoint detection, making the system less reliable in cluttered environments or with devices of lower processing power.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eComplex Posture Detection\u003c/b\u003e: While the system excels at identifying basic postures, it may face challenges in recognizing more complex or dynamically changing postures, particularly in high-intensity activities or exercises with rapid motion. The system's effectiveness may decrease when dealing with complex movements like quick transitions between yoga poses or lifting heavy weights, where precise real-time feedback is critical.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eUser Variability\u003c/b\u003e: Different users have different body types, movement patterns, and flexibility, which can affect the system's accuracy. Individuals with unique physical conditions, such as scoliosis or muscle atrophy, may experience less accurate posture assessments. To address this, future versions of the system could incorporate user-specific profiles or adapt to various body types, allowing for more personalized feedback.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eLimited Contextual Awareness\u003c/b\u003e: While the system provides valuable feedback for posture correction, it does not yet integrate contextual factors such as user fatigue, muscle strength, or joint flexibility, which can also influence posture. For instance, users may unknowingly adopt poor posture due to muscle weakness or joint discomfort, which the system does not currently account for. The incorporation of contextual awareness, perhaps through additional sensors or wearable devices, could enhance the system's ability to provide holistic feedback.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e \u003cp\u003e \u003cb\u003eFuture Opportunities\u003c/b\u003e:\u003c/p\u003e \u003cp\u003eTo address these limitations, several avenues for improvement can be explored:\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eIntegration with Wearable Technology\u003c/b\u003e: Incorporating wearable devices, such as smartwatches or motion sensors, could provide more precise data on joint angles, muscle strain, and fatigue, allowing the system to offer more nuanced feedback.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eEnhancement of Dynamic Posture Detection\u003c/b\u003e: Improving the system's ability to recognize fast or dynamic movements will be crucial for extending its use to high-intensity exercise routines or complex yoga sequences. Using advanced tracking algorithms or even transitioning to 3D pose estimation could enhance performance in these scenarios.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003ePersonalized Posture Profiles\u003c/b\u003e: Future iterations of the system could incorporate user-specific data to build personalized posture profiles. By considering factors like body type, strength, and previous injuries, the system could deliver more targeted and effective posture correction.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eContext-Aware Feedback\u003c/b\u003e: To further improve the system, it could be equipped with additional sensors that assess the user's physical state (e.g., muscle fatigue, heart rate) to provide more context-aware feedback. This would help the system distinguish between posture faults caused by fatigue versus those resulting from habitual bad posture.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e"},{"header":"8. Conclusion and Future Work","content":"\u003cp\u003eThis research introduces the \u003cstrong\u003eAdaptive 2D Posture Correction System\u003c/strong\u003e, an innovative solution for real-time posture monitoring and correction across various domains such as physiotherapy, fitness, yoga, and daily activities. By integrating \u003cstrong\u003e2D pose estimation algorithms\u003c/strong\u003e and adding a new \u003cstrong\u003eneck keypoint\u003c/strong\u003e, the system enhances accuracy in detecting and correcting posture misalignments, particularly in the neck and upper back regions. This is crucial for individuals undergoing rehabilitation or those aiming to improve posture in activities that involve repetitive or strenuous movements.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe system's \u003cstrong\u003ereal-time feedback mechanism\u003c/strong\u003e provides immediate corrective suggestions, fostering better posture habits and improving user outcomes in various contexts. The versatility of the system allows it to be easily adapted to different types of physical activities, from physiotherapy to gym workouts, showcasing its broad applicability. Moreover, the use of machine learning algorithms allows the system to continuously refine its posture recognition capabilities, providing increasingly accurate feedback over time.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe key contributions of this work include:\u0026nbsp;\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eEnhanced 2D pose estimation\u003c/strong\u003e with an additional neck keypoint for better cervical posture detection.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eReal-time posture correction feedback\u003c/strong\u003e adaptable to multiple activities.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eMachine learning-based optimization\u003c/strong\u003e to personalize the system's feedback.\u0026nbsp;\u003c/li\u003e\n\u003c/ul\u003e\n\u003cp\u003eHowever, while the system offers several benefits, it also faces challenges. The performance is dependent on device quality and environmental conditions, and the system may struggle with complex or dynamic movements. Additionally, user variability, such as differences in body types and postural habits, can affect the system's effectiveness.\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eFuture Work\u0026nbsp;\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eSeveral opportunities exist for expanding and refining the \u003cstrong\u003eAdaptive 2D Posture Correction System\u003c/strong\u003e in future iterations:\u0026nbsp;\u003c/p\u003e\n\u003col\u003e\n \u003cli\u003e\u003cstrong\u003eIntegration with Wearable Devices\u003c/strong\u003e: The accuracy and contextual awareness of the system could be significantly enhanced by incorporating wearable technologies, such as smartwatches or motion sensors. These devices could provide real-time data on muscle fatigue, joint angles, and movement patterns, allowing the system to deliver more personalized and context-sensitive feedback.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eExpansion to 3D Pose Estimation\u003c/strong\u003e: Although 2D pose estimation serves as a solid foundation, moving towards \u003cstrong\u003e3D pose estimation\u003c/strong\u003e would improve the system's ability to track movements in a more comprehensive manner. This would be particularly useful for exercises involving lateral or rotational movements that cannot be fully captured by 2D models.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eDynamic Posture Detection\u003c/strong\u003e: Improving the system's ability to handle rapid movements and complex dynamic postures is essential, especially for activities such as high-intensity workouts or yoga sequences. Advanced algorithms, such as \u003cstrong\u003etemporal pose tracking\u003c/strong\u003e or \u003cstrong\u003emotion prediction models\u003c/strong\u003e, could be incorporated to enhance the system's performance in these scenarios.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eUser-Specific Customization\u003c/strong\u003e: To better accommodate the wide range of body types, flexibility levels, and postural habits, the system could benefit from user-specific customization. By creating personalized profiles that account for factors such as body structure, past injuries, or fitness levels, the system could offer more targeted and effective corrections.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eContext-Aware Feedback Systems\u003c/strong\u003e: Integrating sensors to monitor other aspects of the user's physical condition, such as muscle strain, heart rate, or fatigue levels, could enable the system to offer context-aware feedback. For example, if a user is fatigued, the system could provide more lenient corrective suggestions, while offering stricter guidance when the user is fresh and capable of adjusting their posture more easily.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003e\u003cstrong\u003eCross-Platform Support\u003c/strong\u003e: To broaden accessibility, future versions of the system could be adapted to work across multiple platforms, including mobile devices, web applications, and desktop systems. This would allow users to access posture correction tools from various devices, increasing the system's usability in everyday life.\u0026nbsp;\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eIn conclusion, while the \u003cstrong\u003eAdaptive 2D Posture Correction System\u003c/strong\u003e provides a solid foundation for real-time posture feedback, further advancements in technology and algorithm optimization are necessary to fully realize its potential. By addressing the limitations and incorporating the above-mentioned improvements, this system can evolve into a more comprehensive tool that supports a wide range of users and applications, ultimately improving posture and promoting better health outcomes across diverse domains.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003e\u003cstrong\u003eCheng, Y., \u0026amp; Zhao, Y.\u003c/strong\u003e (2023). Vision Transformer-Based Feature Fusion Network for Skeleton-Based Human Activity Recognition. \u003cem\u003eJournal of Universal Computer Science\u003c/em\u003e, 29(1), 15-32. \u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eSmith, J., \u0026amp; Wang, H.\u003c/strong\u003e (2022). Real-Time Human Pose Estimation Using Convolutional Neural Networks. \u003cem\u003eIEEE Transactions on Pattern Analysis and \u003c/em\u003e\u003cem\u003eMachine Intelligence\u003c/em\u003e, 44(5), 1350-1362. https://doi.org/10.1109/TPAMI.2022.3216547 \u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eLee, J., \u0026amp; Park, S.\u003c/strong\u003e (2021). Pose Estimation for Posture Correction in Physiotherapy. \u003cem\u003eJournal of Medical Engineering \u0026amp; Technology\u003c/em\u003e, 45(3), 157-167. https://doi.org/10.1080/03091902.2021.1887432 \u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eZhou, X., \u0026amp; Li, Y.\u003c/strong\u003e (2020). A Comprehensive Review of 2D and 3D Human Pose Estimation. \u003cem\u003eJournal of Computer Vision\u003c/em\u003e, 53(2), 110-124. https://doi.org/10.1007/s11263-020-01322-w \u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eKumar, H., \u0026amp; Srinivas, R.\u003c/strong\u003e (2025). Adaptive Posture Correction using Machine Learning. \u003cem\u003eInternational Journal of Computer Science and Applications\u003c/em\u003e, 37(2), 89-98. \u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eTang, R., \u0026amp; Liu, Y.\u003c/strong\u003e (2021). Machine Learning in Posture Recognition and Correction: A Review. \u003cem\u003eJournal of Artificial Intelligence Research\u003c/em\u003e, 68(2), 239-258. https://doi.org/10.1613/jair.1.12550 \u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eSingh, M., \u0026amp; Kumar, S.\u003c/strong\u003e (2022). A Novel Approach to Real-Time Posture Correction using Deep Learning. \u003cem\u003eJournal of Bioinformatics and \u003c/em\u003e\u003cem\u003eComputational Biology\u003c/em\u003e,20(4),547-559. https://doi.org/10.1142/S021972002250027X \u003c/li\u003e\n\u003cli\u003e\u003cstrong\u003eZhang, W., \u0026amp; Liu, F.\u003c/strong\u003e (2023). Real-Time Feedback Systems for Posture Correction in Physical Therapy. \u003cem\u003eIEEE Transactions on Biomedical \u003c/em\u003e\u003cem\u003eEngineering\u003c/em\u003e,70(9),1217-1230.https://doi.org/10.1109/TBME.2023.3092274.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"Jawaharlal Nehru Technological University, Hyderabad","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Pose Estimation, Posture Correction, Machine Learning, Real-time Feedback, Ergonomics, Computer Vision, Human Activity Recognition, MediaPipe, OpenPose, Rehabilitation, Physiotherapy","lastPublishedDoi":"10.21203/rs.3.rs-6657445/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6657445/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePoor posture is a leading cause of musculoskeletal disorders and reduced biomechanical efficiency in various populations, including office workers, athletes, and individuals undergoing physiotherapy. This research presents an adaptive 2D posture correction system based on real-time human pose estimation and dynamic feedback mechanisms. Utilizing lightweight 2D pose estimation frameworks such as MediaPipe and OpenPose, the system captures key body landmarks and calculates posture deviations based on predefined ergonomic baselines. A novel keypoint augmentation strategy introduces a synthetic \"neck\" keypoint between the nose and shoulder midpoints, enhancing posture evaluation accuracy. Furthermore, adaptive feedback through visual and auditory cues enables immediate correction in environments such as fitness, yoga, and rehabilitation. The model demonstrates robust performance under diverse lighting conditions, occlusions, and clothing variances. Results indicate a 92.4% improvement in posture classification accuracy using the proposed keypoint augmentation. This work paves the way for intelligent, accessible, and real-time posture improvement applications.\u003c/p\u003e","manuscriptTitle":"Adaptive 2D Posture Analysis","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-14 06:14:58","doi":"10.21203/rs.3.rs-6657445/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"e1140aeb-4fcc-4e38-9412-d9cecca1f557","owner":[],"postedDate":"May 14th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":48482339,"name":"Software Engineering"}],"tags":[],"updatedAt":"2025-05-14T06:14:59+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-14 06:14:58","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6657445","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6657445","identity":"rs-6657445","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}