Integrating Data Mining Methods into Smart Drone Systems for Supporting Visually Impaired Users

Integrating Data Mining Methods into Smart Drone Systems for Supporting Visually Impaired Users | 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 Integrating Data Mining Methods into Smart Drone Systems for Supporting Visually Impaired Users Bassam Sabri This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9039478/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 Uncertain outdoor travel in dynamic environments continues to be a problem experienced by the visually impaired. While existing assistive technologies offer valuable assistance, much of it is distance-based and may issue alerts which are too late or lack clarity. In this paper, we propose an augmented information system that integrates smart drone sensing and data-driven analysis for enhanced navigation support outside. A small unmanned aerial vehicle acquires visual, depth, and/or positioning data and transmits the captured data to a processing unit located at an edge. From there, the data are processed and analyzed through methods of clustering, classification, and anomaly detection to learn possible dangerous experiences as well as paths in which safe movements occur. The collected data is presented in a non-intrusive way through brief audio and haptic outputs that minimize cognitive effort. The results of the scenario based simulations demonstrate that compared with the traditional Sensor mounted in ground level, (1) the proposed system is able to recognize outdoor obstacles efficiently; (2) has a better response behavior and (3) yields fewer false warnings. The research provides an information systems perspective on drone assisted navigation, emphasizing the importance of governance, privacy-awareness design and explainability, towards trustworthy decision-support solutions. smart drones data mining assistive information systems visually impaired users navigation decision support Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Highlights • Demonstrates an information tool that connects drone-based sensing with data analysis for assisting outdoor mobility. • Real-time analysis flow for clustering, classification and anomaly detection to evaluate navigation risk. • Demonstrates better obstacle detection performance with improved false positive rates compared to a traditional ground-based method. • Includes governance mechanisms such as privacy constraints, activity logging and explainability to enhance the accountability of the system. • Provides design recommendations for development of an inclusive and robust decision support in safety-critical assistive applications. 1. Introduction IS have come to constitute the debate around people's understanding of, and behaviour in complex environments. '' Blind pedestrian access to timely and clear information about surrounding hazards, walkable space, available movement options is essential in order to navigate independently. Traditional mobility aids such as the white cane and guide dog are still highly effective in many situations; however, they are fundamentally constrained to close-range direct physical interaction. These tools therefore offer little early warning of dynamic circumstances. Thus, more recent works in this direction have investigated the application of digital travel aids built on top of wearable technology and computer vision. Nevertheless, in cluttered outdoor environments where occlusion, dense pedestrian activity and unpredictable motion are present, many of these systems degrade their performance1. A different sensing point of view can be searched for thanks to smart drones that, from the above, are able to detect obstacles and movement (dance-like) patterns. Nevertheless, UAV platforms produce heterogeneous data streams that are not useful for navigation without proper processing. The use of data mining such as clustering, classification, and anomaly detection provides a practical means of extracting patterns that are pertinent to these phenomena from the large amounts of data available and converting them into actionable knowledge for decision support [2,3]. From an IS perspective, the issue of concern is not only that of accurate perception, but also pertains to whether the information delivered is reliable; transparent; and subject to governance, control and trust within systems that are operating under safety-critical work scenarios [4,5]. In this context, the research explores: “How smart drone technologies supported with data mining techniques can be employed to assist visually impaired users in an assistance system?” The paper presents four types of contributions: (i) a layered architecture of IS for drone-assisted navigation; (ii) an analytics pipeline interconnecting perception with risk assessment and alert prioritization pathways; (iii) scenario-based evaluation approach, available in on-going European Research Project project DREAMS 4.0 as well as a testbed to evaluate the above-mentioned architecture and design pipelines; and (iv) placing privacy protection, accountability, and explainability at the fore-front of system design implications. 2. Background and Related Work Assistive navigation has been addressed by a number of works, including assistive devices for people with visual impairments, vision-based smartphone applications and infrastructure-supported systems. Previous review studies often discuss a trade-off between how far the system can sense, how trustworthy the system can be and how much mental activity is needed from the user. the relationship between them can be described by the following utility function:: U = αR + βρ − γL (1) where R, ρ and L are sensing range, reliability and cognitive load [1], respectively. Although vision-based approaches can offer precise scene information, they tend to perform poorly in practical outdoor conditions. For instance, poor lighting, camera movement and occlusion hinder the correct recognition chance (see Fig. 1 .) as reported in [6]. UAVs have been studied as a method of increasing situation awareness, and guidance using auditory and haptic feedback. Experimental evidence indicates that people are able to effectively move behind a drone as visual or directional reference, if interaction rules and safety limits are properly enforced [7–9]. For a navigation safety perspective, guidance can be characterized by imposing the minimum permissible distance between the user and obstacles. d(t)≥dmin​, (2) whilst maintaining a movement within a predetermined safe trajectory. This constraint is illustrated in Fig. 2 ., which depicts situational awareness at a road convergent point. When information systems deal with data that represent some aspect of the human context, concerns about responsible use, transparency, supervision are exceedingly crucial [4,5,10]. These worries are the ones that drive practical limits to data retention time and where computation is done. Studies on real-time UAV based analytics also confirm that processing is often offloaded to the edge, and lightweight analytical models are preferred due to tight time deadlines and limited energy capacity [11,12]. Therefore, the system behavior also has to be kept within some bounded end-to-end response time. Te2e≤Tmax (3) At the same time, energy consumption on the drone needs to remain in the realm of practicality. This restriction promotes on-device data processing/analysis as opposed to overpowering cloud-based computations. Within this context, UAVs have been addressed as devices to enhance situational awareness and to assist navigation with audio and haptic cues. An example for this design is illustrated in Fig. 3 . Preliminary studies indicate that users can navigate by tracking a lead drone as long as safety boundaries and interaction design parameters are adhered to [7–9]. In parallel, studies in information systems highlight how systems process sensitive and context-dependent personal human data, underscoring the requirements for transparent governance [4], [5], [10]. Similarly, the study of real-time UAV analytics also remarks that the power and processing constraints demand edge-based light-weighted calculations, as opposed to computationally expensive techniques [11,12]. 3. Research Model and System Architecture The system is divided into functional layers as illustrated in Fig. 4 . The UAV collects sensor information from its local area and forwards them to an edge entity, which can be hosted either on a drone or in a tethered smartphone. This member computes the data and provides feedback about the navigation to the user. In support of responsible operation, the system is designed with a governance layer that controls privacy settings, logs system activity and provides minimal explanations for system decisions to build trust and accountability [4,5]. The main data mining components and their corresponding decision-support outputs are summarized in Table 1 . Table 1 Components of the data mining and outputs. Mining component Operational role Decision-support output Clustering Partition local scene into free space and obstacle regions Traversable corridor + confidence Classification Label obstacle type (pedestrian/vehicle/static) under edge constraints Semantic label + distance Anomaly detection Flag unusual motion or unexpected hazards Priority alert + risk score 4. Methods The system should be robust enough to support outdoor movement on sidewalks and crossing pedestrian crossings. The color images and depth data are obtained by the UAV, while its position and orientation are provided by GPS/inertial sensors. Received data are first temporally aligned and then filtered for noise. Thereby basic features are derived, such as the location and distance of objects, their moving speed, and estimates of available walking space. These characteristics are dealt with as continuous flows so that the analysis can be done with a low latency. 4.1 Clustering for identifying walkable areas Clustering then groups depth points that are closeby and opens up walking areas from cluttered/surrounding regions. To avoid spurious guidance, the clustering results are combined with evidence of sensor quality (i.e. depth measurements fluctuations when strong ambient environment occurred, as problematic situation). 4.2 Classification for environmental awareness A lightweight classification model is used to locate common scene elements, e.g. pedestrians, vehicles and fixed objects. By using such labels, the system can produce more explicit guidance messages (e.g. separating into static and moving obstacles) resulting in less confusion compared to generic alerts [6]. 4.3 Anomaly detection for risk handling Anomaly detection is concerned with separation of the unusual motion behavior (or an unexpected trajectory change). Instances where objects are moving in and out very quickly (like people crossing the street with their cellphones) or there's a sudden change in pedestrian flow are considered higher risk. Events are detected in a hierarchical manner so that important warnings are presented sooner than unimportant ones to prevent unnecessary disturbance to the user and cognitive effort. The evaluation scenarios considered in this study, along with their primary hazards, are summarized in Table 2 . Table 2 Evaluation scenarios and primary hazards. Scenario Description Primary hazards S1 Routine Sidewalk with poles, curbs, signs Static obstacles; narrow passages S2 Accidental entry Pedestrians intermittently enter corridor Near-collisions; occlusions S3 Vehicle approach Crossing with approaching vehicles Fast objects; speed estimation S4 Coordinated hazards Multiple agents + sudden blockage Compound risks; alert prioritization 5. Results Performance of the system is evaluated based on three features: correctness of obstacle detection, false warning ratio and reaction time. Our results are compared with the ground-based reference approach where short-range sensing is used, and alert rules are predefined according to classifications often seen in assistive navigation literature [1] Fig. 5 . compares the overall performance of the proposed UAV-based system with the ground-based baseline in terms of detection accuracy, false alerts, and response time. As shown in Fig. 6 ., the end-to-end latency increases with scenario complexity, while remaining within acceptable response-time limits. Aggregate performance results across all scenarios are reported in Table 3 . Table 3 Aggregate performance across scenarios (mean values). Metric Ground-based baseline Proposed UAV + data mining Obstacle detection accuracy (%) 78.5 91.2 False alert rate (%) 15.3 6.4 Average response time (s) 1.8 1.1 In all scenarios studied the aerial sensing can let the user know about an obstacle sooner, than are not directly in his close range. Meanwhile, the data processing pipeline refines warnings by rejecting ambiguous signals and concentrating on higher-risk events. We are able to achieve good class-wise separation among the primary scene classes under normal conditions This behavior is reflected in the normalized confusion matrix shown in Fig. 7 ., which indicates good class-wise separation under normal conditions for being more context-sensitive. 6. Discussion The results suggest that the integration of drone-based sensing and data-driven analysis may accelerate delivery and enhance reliability of mobility help. From an information systems entity POV, the value of such a system is critically dependent on the quality of the available information, especially in terms of accuracy, reliability and timelyness - within cognitive and safety limits. The governance factor becomes critical due to the fact that in using drones we inevitably capture nearby individuals, which can be controlled partially by restricting data collection and embedding privacy-by-design, ultimately treating ethical and regulatory implications (cf. [4,5,10]). Several challenges remain in practice. Battery size restrictions the operation period, regulatory guidelines can prevent deployment in populated areas and user acceptance is to a large extend connected with perceived comfort and safety. Future research should therefore have its focus on real life testing with user’s centered UI design and investigate how clear, understandable risk information affects trust and continued use. 7. Implications for IS Research and Practice From a research point of view, this work demonstrates how drone based sensing can be considered as a decision-support information system that accommodating real-time data processing, analytical operations and user orientated interaction. In the real world, the presented architecture provides a roadmap for developing assistive systems that focus on actionable rather than raw sensor data with explicit consideration of privacy by design and also system accountability. 8. Conclusion This study developed an assistive information system with smart drones and data-driven methods to support navigation of visually impaired people. Clustering, classification, and anomaly detection of aerial sensor data are performed in order to produce guidance based on risk as it is disseminated via audio and haptic notifications. Simulations demonstrate superior obstacle detection, less unnecessary warnings and faster reaction times with respect to a terrestrial reference system. They will continue to advance efforts to evaluate in real-world settings, and to personalize the tool for individual users, as well as to create governance structures that would foster broader safe and responsible use. Declarations Ethics Statement This study does not involve experiments on human participants or animals. The work is based on system design, simulation, and scenario-based evaluation. No personal or identifiable data were collected, stored, or processed. Therefore, formal ethical approval was not required. Funding The author received no financial support for the research, authorship, and/or publication of this article. Declaration of competing interest The authors have no conflicts of interest to report concerning the work in this paper. References Dakopoulos, D., & Bourbakis, N. G. (2010). Wearable obstacle avoidance electronic travel aids for blind: A survey. IEEE Transactions on Systems, Man, and Cybernetics—Part C, 40(1), 25–35. https://doi.org/10.1109/TSMCC.2009.2022281 Chen, H., Chiang, R. H. L., & Storey, V. C. (2012). Business intelligence and analytics: From big data to big impact. MIS Quarterly, 36(4), 1165–1188. Aggarwal, C. C. (2015). Data mining: The textbook. Springer. Dwivedi, Y. K., et al. (2020). Impact of emerging technologies on information systems research: A position paper. Information Systems Journal, 30(3), 400–420. Sarker, I. H. (2021). Machine learning: Algorithms, real-world applications and research directions. SN Computer Science, 2, 160. Katzschmann, R. K., Araki, B., & Rus, D. (2018). Safe local navigation for visually impaired users with a time-of-flight and RGB camera. IEEE Robotics and Automation Letters, 3(4), 3376–3383. https://doi.org/10.1109/LRA.2018.2833999 Cauchard, J. R., Zhai, K. Y., Spadafora, M., & Landay, J. A. (2015). Emotion encoding in human-drone interaction. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 1(3), 1–22. https://doi.org/10.1145/2803164 Scheible, J., & Sakamoto, D. (2019). Drone-guided navigation support for visually impaired people in outdoor environments. International Journal of Human–Computer Studies, 130, 1–14. https://doi.org/10.1016/j.ijhcs.2019.01.002 Azenkot, S., et al. (2011). Overcoming challenges of non-visual public transit use. CHI Proceedings, 1–10. Shneiderman, B. (2020). Human-centered AI. Oxford University Press. Floreano, D., & Wood, R. J. (2015). Science, technology and the future of small autonomous drones. Nature, 521, 460–466. https://doi.org/10.1038/nature14590 Zeng, Y., Zhang, R., & Lim, T. J. (2016). Wireless communications with unmanned aerial vehicles: Opportunities and challenges. IEEE Communications Magazine, 54(5), 36–42. https://doi.org/10.1109/MCOM.2016.7470909 Chandola, V., Banerjee, A., & Kumar, V. (2009). Anomaly detection: A survey. ACM Computing Surveys, 41(3), 1–58. https://doi.org/10.1145/1541880.1541884 Han, J., Kamber, M., & Pei, J. (2012). Data mining: Concepts and techniques (3rd ed.). Morgan Kaufmann. Wamba, S. F., et al. (2017). Big data analytics and firm performance: Effects of dynamic capabilities. Journal of Business Research, 70, 356–365. https://doi.org/10.1016/j.jbusres.2016.08.016 Floridi, L., et al. (2018). AI4People—An ethical framework for a good AI society. Minds and Machines, 28, 689–707. https://doi.org/10.1007/s11023-018-9482-3 Nissenbaum, H. (2010). Privacy in context: Technology, policy, and the integrity of social life. Stanford University Press. ISO/IEC 27001:2022. (2022). Information security, cybersecurity and privacy protection—formation security management systems—Requirements. Raj, A., & Seamans, R. (2019). Primer on AI and IS research. MIS Quarterly, 43(3), iii–viii. https://doi.org/10.25300/MISQ/2019/43.3 Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). 'Why should I trust you?': Explaining the predictions of any classifier. KDD Proceedings, 1135–1144. https://doi.org/10.1145/2939672.2939778. Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterialCode1.docx SupplementaryResults.docx SupplementaryTables.docx 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-9039478","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":630829233,"identity":"7955881d-927b-4e8d-b43c-51678764b439","order_by":0,"name":"Bassam Sabri","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA9ElEQVRIiWNgGAWjYBACAxiDj5n5AIxrgCSORwsbM1sCkGtAihYGHgMYH496IDBnP/74w88cBnk2dp5vEh8K/sgzsDdvk2AosMapxbInx0yydxuDYRsz7zbJGQYGhg08x8okGAzScTvsQA4bA+82BkaQFmkeA4MEBokcM6CWw7i1nH/++OPfbQz2bcw8zyBa5N8Q0HIjwUAaaEsiUAsb1BYe/FosZ7wxk5bdJpHcxsxmbDnDwNiwjSet2CIBj1/M+dMff3y7zca2n//wwxsf/sjJ87Mf3ghk4A4xKJBAMNlARAIDMyEtmIAMLaNgFIyCUTBcAQAn/UMABP6s6AAAAABJRU5ErkJggg==","orcid":"","institution":"University of Information Technology and Communications","correspondingAuthor":true,"prefix":"","firstName":"Bassam","middleName":"","lastName":"Sabri","suffix":""}],"badges":[],"createdAt":"2026-03-05 11:24:41","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9039478/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9039478/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108491147,"identity":"c13aa8c9-9b98-417e-baf0-c18fda7e13a1","added_by":"auto","created_at":"2026-05-05 09:52:31","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":317821,"visible":true,"origin":"","legend":"\u003cp\u003eillustrates the overall data flow architecture of the proposed smart drone–based assistive navigation system.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/47f20fdad481978c480a0c3c.png"},{"id":108491667,"identity":"30061b63-a638-438c-adb8-d1e114bcd984","added_by":"auto","created_at":"2026-05-05 09:55:08","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":484995,"visible":true,"origin":"","legend":"\u003cp\u003eSituational awareness at a road convergent point with UAV.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/5f56954786150e1486594e67.png"},{"id":108229061,"identity":"1e7eb372-4b3b-4521-b7b5-3f0985caf637","added_by":"auto","created_at":"2026-04-30 17:01:01","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":202915,"visible":true,"origin":"","legend":"\u003cp\u003eillustrates the task offloading and edge-assisted data processing model adopted by the system.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/78bace67aa13e61ba8f54f8a.png"},{"id":108229062,"identity":"a3664f63-4350-4765-8bfc-e9d2c48b4e2b","added_by":"auto","created_at":"2026-04-30 17:01:01","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":68398,"visible":true,"origin":"","legend":"\u003cp\u003eThe architecture of the assistive information system based on UAV.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/8384f8a795344d80384dfebf.png"},{"id":109202993,"identity":"509f73cf-ae56-4116-9b53-53e2bf75ebdd","added_by":"auto","created_at":"2026-05-13 14:20:13","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":39011,"visible":true,"origin":"","legend":"\u003cp\u003eOverall performance comparison between the ground-based baseline and the proposed UAV + data mining system.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/e06bf2a008f8192bb4b0043d.png"},{"id":108229065,"identity":"2699fda2-6d4d-4829-b430-737736dac385","added_by":"auto","created_at":"2026-04-30 17:01:01","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":59599,"visible":true,"origin":"","legend":"\u003cp\u003eEnd-to-end latency versus scenario complexity (higher complexity implies more obstacles and moving agents).\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/57cd1038a05ad33271f45b67.png"},{"id":108491911,"identity":"554f9c59-72be-4db4-9ce8-c181f9cd31e7","added_by":"auto","created_at":"2026-05-05 09:56:10","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":73849,"visible":true,"origin":"","legend":"\u003cp\u003eClassification performance (normalized confusion matrix) for four scene classes (synthetic illustration).\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/9aa38d6f30b61150e477d5f8.png"},{"id":109375523,"identity":"28427983-a7c2-4344-b076-2e577e70d5d8","added_by":"auto","created_at":"2026-05-16 14:25:07","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1277093,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/2bdf4399-ed16-4021-ac68-98be24935a10.pdf"},{"id":108229057,"identity":"e4bc4d6f-387e-486d-a96b-53cb80e713c4","added_by":"auto","created_at":"2026-04-30 17:01:01","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":33237,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterialCode1.docx","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/642a7402dc31a50d7b6a3b4e.docx"},{"id":108229059,"identity":"4d1fdd5a-389e-44a4-a8f9-2e26ee092d6d","added_by":"auto","created_at":"2026-04-30 17:01:01","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":27707,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryResults.docx","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/0a6d76ca476cbe9f254fc578.docx"},{"id":108229060,"identity":"26f9dd69-28dc-41c6-8ed8-dc21856b09dc","added_by":"auto","created_at":"2026-04-30 17:01:01","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":27141,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-9039478/v1/91866d40944a3441983d0925.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Integrating Data Mining Methods into Smart Drone Systems for Supporting Visually Impaired Users","fulltext":[{"header":"Highlights","content":"\u003cp\u003e\u0026bull; Demonstrates an information tool that connects drone-based sensing with\u0026ensp;data analysis for assisting outdoor mobility.\u003c/p\u003e\u003cp\u003e\u0026bull; Real-time analysis flow for clustering, classification\u0026ensp;and anomaly detection to evaluate navigation risk.\u003c/p\u003e\u003cp\u003e\u0026bull; Demonstrates better obstacle detection performance with improved false positive rates\u0026ensp;compared to a traditional ground-based method.\u003c/p\u003e\u003cp\u003e\u0026bull; Includes governance mechanisms such as privacy constraints, activity logging and explainability to enhance the accountability\u0026ensp;of the system.\u003c/p\u003e\u003cp\u003e\u0026bull; Provides\u0026ensp;design recommendations for development of an inclusive and robust decision support in safety-critical assistive applications.\u003c/p\u003e"},{"header":"1. Introduction","content":"\u003cp\u003eIS have come to constitute the\u0026ensp;debate around people\u0026apos;s understanding of, and behaviour in complex environments. \u0026apos;\u0026apos; Blind pedestrian access to timely and clear information about surrounding hazards, walkable space, available movement\u0026ensp;options is essential in order to navigate independently. Traditional mobility aids such as the white cane and guide dog\u0026ensp;are still highly effective in many situations; however, they are fundamentally constrained to close-range direct physical interaction. These tools therefore offer little early warning of dynamic\u0026ensp;circumstances. Thus, more recent works in this direction have investigated the application of digital travel aids built\u0026ensp;on top of wearable technology and computer vision. Nevertheless, in cluttered outdoor environments\u0026ensp;where occlusion, dense pedestrian activity and unpredictable motion are present, many of these systems degrade their performance1.\u003c/p\u003e\n\u003cp\u003eA different sensing point of view can be searched for thanks\u0026ensp;to smart drones that, from the above, are able to detect obstacles and movement (dance-like) patterns. Nevertheless, UAV platforms produce heterogeneous data streams that are not useful\u0026ensp;for navigation without proper processing. The use of data mining such\u0026ensp;as clustering, classification, and anomaly detection provides a practical means of extracting patterns that are pertinent to these phenomena from the large amounts of data available and converting them into actionable knowledge for decision support [2,3]. From an IS perspective, the issue of concern is not only that of accurate perception, but also pertains to whether the information\u0026ensp;delivered is reliable; transparent; and subject to governance, control and trust within systems that are operating under safety-critical work scenarios [4,5].\u003c/p\u003e\n\u003cp\u003eIn this context, the research explores: \u0026ldquo;How smart drone technologies supported with data mining techniques\u0026ensp;can be employed to assist visually impaired users in an assistance system?\u0026rdquo; The paper presents four types of contributions: (i) a layered architecture of IS for drone-assisted navigation; (ii) an analytics\u0026ensp;pipeline interconnecting perception with risk assessment and alert prioritization pathways; (iii) scenario-based evaluation approach, available in on-going European Research Project project DREAMS 4.0 as well as a testbed to evaluate the above-mentioned architecture and design pipelines; and (iv) placing privacy protection, accountability, and explainability at the fore-front of system design implications.\u003c/p\u003e"},{"header":"2. Background and Related Work","content":"\u003cp\u003eAssistive navigation has been addressed by a number of\u0026ensp;works, including assistive devices for people with visual impairments, vision-based smartphone applications and infrastructure-supported systems. Previous review studies often discuss a trade-off between how far the system can sense, how trustworthy the system can be and how much mental activity is needed\u0026ensp;from the user. the relationship between them\u0026ensp;can be described by the following utility function::\u003c/p\u003e \u003cp\u003eU\u0026thinsp;=\u0026thinsp;αR\u0026thinsp;+\u0026thinsp;βρ\u0026thinsp;\u0026minus;\u0026thinsp;γL (1)\u003c/p\u003e \u003cp\u003ewhere R, ρ and L are sensing range, reliability and cognitive load [1],\u0026ensp;respectively. Although vision-based approaches can offer precise scene information, they tend to\u0026ensp;perform poorly in practical outdoor conditions. For instance, poor lighting, camera movement and occlusion hinder\u0026ensp;the correct recognition chance (see Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.) as reported in [6].\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eUAVs have been studied as a method of increasing situation awareness, and\u0026ensp;guidance using auditory and haptic feedback. Experimental evidence indicates that people are able to effectively move behind a drone as\u0026ensp;visual or directional reference, if interaction rules and safety limits are properly enforced [7\u0026ndash;9]. For\u0026ensp;a navigation safety perspective, guidance can be characterized by imposing the minimum permissible distance between the user and obstacles.\u003c/p\u003e \u003cp\u003ed(t)\u0026ge;dmin​, (2)\u003c/p\u003e \u003cp\u003ewhilst maintaining a movement within a predetermined\u0026ensp;safe trajectory. This constraint is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e., which depicts situational awareness at a road convergent point.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eWhen information systems deal with data that represent\u0026ensp;some aspect of the human context, concerns about responsible use, transparency, supervision are exceedingly crucial [4,5,10]. These worries are the ones that drive practical limits to\u0026ensp;data retention time and where computation is done. Studies on real-time UAV based analytics also confirm that processing is often offloaded to the edge, and lightweight analytical models are preferred due\u0026ensp;to tight time deadlines and limited energy capacity [11,12]. Therefore, the system\u0026ensp;behavior also has to be kept within some bounded end-to-end response time.\u003c/p\u003e \u003cp\u003eTe2e\u0026le;Tmax (3)\u003c/p\u003e \u003cp\u003eAt the same time, energy consumption on the\u0026ensp;drone needs to remain in the realm of practicality. This restriction promotes on-device data processing/analysis\u0026ensp;as opposed to overpowering cloud-based computations. Within this context, UAVs have been addressed as devices to enhance situational awareness and to assist navigation with\u0026ensp;audio and haptic cues. An example for this design is\u0026ensp;illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003ePreliminary studies indicate that users can navigate by tracking a lead drone as long as safety boundaries\u0026ensp;and interaction design parameters are adhered to [7\u0026ndash;9]. In parallel,\u0026ensp;studies in information systems highlight how systems process sensitive and context-dependent personal human data, underscoring the requirements for transparent governance [4], [5], [10]. Similarly, the study of real-time UAV analytics also remarks that the power and processing constraints demand\u0026ensp;edge-based light-weighted calculations, as opposed to computationally expensive techniques [11,12].\u003c/p\u003e"},{"header":"3. Research Model and System Architecture","content":"\u003cp\u003eThe system is divided into functional layers as illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e. The UAV collects sensor information from its local area and forwards them to an\u0026ensp;edge entity, which can be hosted either on a drone or in a tethered smartphone. This member computes the data and provides feedback about the\u0026ensp;navigation to the user. In support of responsible operation, the system is designed with a governance layer that\u0026ensp;controls privacy settings, logs system activity and provides minimal explanations for system decisions to build trust and accountability [4,5].\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe main data mining components and their corresponding decision-support outputs are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComponents of the\u0026ensp;data mining and outputs.\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=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMining component\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eOperational role\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDecision-support output\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClustering\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePartition local scene into free space and obstacle regions\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTraversable corridor\u0026thinsp;+\u0026thinsp;confidence\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClassification\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLabel obstacle type (pedestrian/vehicle/static) under edge constraints\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSemantic label\u0026thinsp;+\u0026thinsp;distance\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAnomaly detection\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eFlag unusual motion or unexpected hazards\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePriority alert\u0026thinsp;+\u0026thinsp;risk score\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"4. Methods","content":"\u003cp\u003eThe system should be robust\u0026ensp;enough to support outdoor movement on sidewalks and crossing pedestrian crossings. The color images and depth data are obtained by the UAV, while its\u0026ensp;position and orientation are provided by GPS/inertial sensors. Received data\u0026ensp;are first temporally aligned and then filtered for noise. Thereby basic features are derived, such as the location and distance of objects, their moving speed,\u0026ensp;and estimates of available walking space. These characteristics are dealt with as continuous flows so that the analysis can be done\u0026ensp;with a low latency.\u003c/p\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e4.1 Clustering for identifying walkable areas\u003c/h2\u003e \u003cp\u003eClustering then groups\u0026ensp;depth points that are closeby and opens up walking areas from cluttered/surrounding regions. To avoid spurious guidance, the clustering results are combined with evidence of sensor quality\u0026ensp;(i.e. depth measurements fluctuations when strong ambient environment occurred, as problematic situation).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e4.2 Classification for environmental awareness\u003c/h2\u003e \u003cp\u003eA lightweight classification model is\u0026ensp;used to locate common scene elements, e.g. pedestrians, vehicles and fixed objects. By using such labels, the system can produce more explicit guidance messages (e.g. separating into static and moving obstacles) resulting in less\u0026ensp;confusion compared to generic alerts [6].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e4.3 Anomaly detection for risk handling\u003c/h2\u003e \u003cp\u003eAnomaly detection is concerned with\u0026ensp;separation of the unusual motion behavior (or an unexpected trajectory change). Instances where objects are moving in\u0026ensp;and out very quickly (like people crossing the street with their cellphones) or there's a sudden change in pedestrian flow are considered higher risk. Events are detected in a hierarchical manner so that important warnings are presented sooner than unimportant ones to prevent unnecessary disturbance to\u0026ensp;the user and cognitive effort. The evaluation scenarios considered in this study, along with their primary hazards, are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eEvaluation scenarios and primary hazards.\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=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eScenario\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eDescription\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrimary hazards\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS1 Routine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eSidewalk with poles, curbs, signs\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eStatic obstacles; narrow passages\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS2 Accidental entry\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePedestrians intermittently enter corridor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eNear-collisions; occlusions\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS3 Vehicle approach\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCrossing with approaching vehicles\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eFast objects; speed estimation\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eS4 Coordinated hazards\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eMultiple agents\u0026thinsp;+\u0026thinsp;sudden blockage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eCompound risks; alert prioritization\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"},{"header":"5. Results","content":"\u003cp\u003ePerformance of the system is evaluated based on three features: correctness of obstacle detection,\u0026ensp;false warning ratio and reaction time. Our results are compared with the ground-based reference\u0026ensp;approach where short-range sensing is used, and alert rules are predefined according to classifications often seen in assistive navigation literature [1] Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e. compares the overall performance of the proposed UAV-based system with the ground-based baseline in terms of detection accuracy, false alerts, and response time.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e., the end-to-end latency increases with scenario complexity, while remaining within acceptable response-time limits.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAggregate performance results across all scenarios are reported in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\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\u003eAggregate performance across scenarios (mean values).\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\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGround-based baseline\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eProposed UAV\u0026thinsp;+\u0026thinsp;data mining\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eObstacle detection accuracy (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e78.5\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFalse alert rate (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e6.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAverage response time (s)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1.8\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.1\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\u003e \u003c/p\u003e \u003cp\u003eIn all scenarios studied the aerial sensing can let the user know about an obstacle sooner, than are not directly\u0026ensp;in his close range. Meanwhile, the data processing pipeline refines warnings by rejecting ambiguous signals and\u0026ensp;concentrating on higher-risk events. We are able to achieve good class-wise separation\u0026ensp;among the primary scene classes under normal conditions This behavior is reflected in the normalized confusion matrix shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e., which indicates good class-wise separation under normal conditions for being more context-sensitive.\u003c/p\u003e"},{"header":"6. Discussion","content":"\u003cp\u003eThe results suggest that the integration of drone-based\u0026ensp;sensing and data-driven analysis may accelerate delivery and enhance reliability of mobility help. From an information systems entity POV, the value of\u0026ensp;such a system is critically dependent on the quality of the available information, especially in terms of accuracy, reliability and timelyness - within cognitive and safety limits. The governance\u0026ensp;factor becomes critical due to the fact that in using drones we inevitably capture nearby individuals, which can be controlled partially by restricting data collection and embedding privacy-by-design, ultimately treating ethical and regulatory implications (cf. [4,5,10]).\u003c/p\u003e \u003cp\u003eSeveral challenges remain in practice. Battery size restrictions the operation period, regulatory guidelines can prevent deployment\u0026ensp;in populated areas and user acceptance is to a large extend connected with perceived comfort and safety. Future research should therefore have its focus on real life testing with user\u0026rsquo;s centered UI design and investigate how\u0026ensp;clear, understandable risk information affects trust and continued use.\u003c/p\u003e"},{"header":"7. Implications for IS Research and Practice","content":"\u003cp\u003eFrom a research point of view, this work demonstrates how drone based sensing can be considered as a decision-support information system that accommodating real-time data processing,\u0026ensp;analytical operations and user orientated interaction. In the real world, the presented architecture\u0026ensp;provides a roadmap for developing assistive systems that focus on actionable rather than raw sensor data with explicit consideration of privacy by design and also system accountability.\u003c/p\u003e"},{"header":"8. Conclusion","content":"\u003cp\u003eThis study developed an assistive information system with smart drones and data-driven methods to support navigation\u0026ensp;of visually impaired people. Clustering, classification, and anomaly detection of aerial sensor data are performed in order to\u0026ensp;produce guidance based on risk as it is disseminated via audio and haptic notifications. Simulations demonstrate superior obstacle detection, less\u0026ensp;unnecessary warnings and faster reaction times with respect to a terrestrial reference system. They will continue to advance efforts to evaluate in real-world settings, and to personalize the tool for individual users, as well as to create governance structures that would foster broader\u0026ensp;safe and responsible use.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study does not involve experiments on human participants or animals. The work is based on system design, simulation, and scenario-based evaluation. No personal or identifiable data were collected, stored, or processed. Therefore, formal ethical approval was not required.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author received no financial support for the research, authorship, and/or publication of this article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of competing interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors have no conflicts of interest to report concerning the work in this paper.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eDakopoulos, D., \u0026amp; Bourbakis, N. G. (2010). Wearable obstacle avoidance electronic travel aids for blind: A survey. IEEE Transactions on Systems, Man, and Cybernetics\u0026mdash;Part C, 40(1), 25\u0026ndash;35. https://doi.org/10.1109/TSMCC.2009.2022281\u003c/li\u003e\n \u003cli\u003eChen, H., Chiang, R. H. L., \u0026amp; Storey, V. C. (2012). Business intelligence and analytics: From big data to big impact. MIS Quarterly, 36(4), 1165\u0026ndash;1188.\u003c/li\u003e\n \u003cli\u003eAggarwal, C. C. (2015). Data mining: The textbook. Springer.\u003c/li\u003e\n \u003cli\u003eDwivedi, Y. K., et al. (2020). Impact of emerging technologies on information systems research: A position paper. Information Systems Journal, 30(3), 400\u0026ndash;420.\u003c/li\u003e\n \u003cli\u003eSarker, I. H. (2021). Machine learning: Algorithms, real-world applications and research directions. SN Computer Science, 2, 160.\u003c/li\u003e\n \u003cli\u003eKatzschmann, R. K., Araki, B., \u0026amp; Rus, D. (2018). Safe local navigation for visually impaired users with a time-of-flight and RGB camera. IEEE Robotics and Automation Letters, 3(4), 3376\u0026ndash;3383. https://doi.org/10.1109/LRA.2018.2833999\u003c/li\u003e\n \u003cli\u003eCauchard, J. R., Zhai, K. Y., Spadafora, M., \u0026amp; Landay, J. A. (2015). Emotion encoding in human-drone interaction. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 1(3), 1\u0026ndash;22. https://doi.org/10.1145/2803164\u003c/li\u003e\n \u003cli\u003eScheible, J., \u0026amp; Sakamoto, D. (2019). Drone-guided navigation support for visually impaired people in outdoor environments. International Journal of Human\u0026ndash;Computer Studies, 130, 1\u0026ndash;14. https://doi.org/10.1016/j.ijhcs.2019.01.002\u003c/li\u003e\n \u003cli\u003eAzenkot, S., et al. (2011). Overcoming challenges of non-visual public transit use. CHI Proceedings, 1\u0026ndash;10.\u003c/li\u003e\n \u003cli\u003eShneiderman, B. (2020). Human-centered AI. Oxford University Press.\u003c/li\u003e\n \u003cli\u003eFloreano, D., \u0026amp; Wood, R. J. (2015). Science, technology and the future of small autonomous drones. Nature, 521, 460\u0026ndash;466. https://doi.org/10.1038/nature14590\u003c/li\u003e\n \u003cli\u003eZeng, Y., Zhang, R., \u0026amp; Lim, T. J. (2016). Wireless communications with unmanned aerial vehicles: Opportunities and challenges. IEEE Communications Magazine, 54(5), 36\u0026ndash;42. https://doi.org/10.1109/MCOM.2016.7470909\u003c/li\u003e\n \u003cli\u003eChandola, V., Banerjee, A., \u0026amp; Kumar, V. (2009). Anomaly detection: A survey. ACM Computing Surveys, 41(3), 1\u0026ndash;58. https://doi.org/10.1145/1541880.1541884\u003c/li\u003e\n \u003cli\u003eHan, J., Kamber, M., \u0026amp; Pei, J. (2012). Data mining: Concepts and techniques (3rd ed.). Morgan Kaufmann.\u003c/li\u003e\n \u003cli\u003eWamba, S. F., et al. (2017). Big data analytics and firm performance: Effects of dynamic capabilities. Journal of Business Research, 70, 356\u0026ndash;365. https://doi.org/10.1016/j.jbusres.2016.08.016\u003c/li\u003e\n \u003cli\u003eFloridi, L., et al. (2018). AI4People\u0026mdash;An ethical framework for a good AI society. Minds and Machines, 28, 689\u0026ndash;707. https://doi.org/10.1007/s11023-018-9482-3\u003c/li\u003e\n \u003cli\u003eNissenbaum, H. (2010). Privacy in context: Technology, policy, and the integrity of social life. Stanford University Press.\u003c/li\u003e\n \u003cli\u003eISO/IEC 27001:2022. (2022). Information security, cybersecurity and privacy protection\u0026mdash;formation security management systems\u0026mdash;Requirements.\u003c/li\u003e\n \u003cli\u003eRaj, A., \u0026amp; Seamans, R. (2019). Primer on AI and IS research. MIS Quarterly, 43(3), iii\u0026ndash;viii. https://doi.org/10.25300/MISQ/2019/43.3\u003c/li\u003e\n \u003cli\u003eRibeiro, M. T., Singh, S., \u0026amp; Guestrin, C. (2016). \u0026apos;Why should I trust you?\u0026apos;: Explaining the predictions of any classifier. KDD Proceedings, 1135\u0026ndash;1144. https://doi.org/10.1145/2939672.2939778.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","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":"smart drones, data mining, assistive information systems, visually impaired users, navigation, decision support","lastPublishedDoi":"10.21203/rs.3.rs-9039478/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9039478/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eUncertain\u0026ensp;outdoor travel in dynamic environments continues to be a problem experienced by the visually impaired. While existing assistive technologies offer valuable assistance, much of it is distance-based\u0026ensp;and may issue alerts which are too late or lack clarity. In this paper, we propose an augmented information system that integrates smart drone sensing and data-driven analysis for enhanced navigation support\u0026ensp;outside. A small unmanned aerial vehicle acquires visual,\u0026ensp;depth, and/or positioning data and transmits the captured data to a processing unit located at an edge. From there, the data are processed and analyzed\u0026ensp;through methods of clustering, classification, and anomaly detection to learn possible dangerous experiences as well as paths in which safe movements occur. The collected data is presented in a non-intrusive way\u0026ensp;through brief audio and haptic outputs that minimize cognitive effort. The results of the scenario based simulations demonstrate that compared with the traditional Sensor mounted in ground level, (1) the proposed system is able to recognize outdoor obstacles efficiently; (2)\u0026ensp;has a better response behavior and (3) yields fewer false warnings. The research provides an information systems perspective on drone assisted navigation, emphasizing the importance of governance, privacy-awareness design and explainability,\u0026ensp;towards trustworthy decision-support solutions.\u003c/p\u003e","manuscriptTitle":"Integrating Data Mining Methods into Smart Drone Systems for Supporting Visually Impaired Users","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-04-30 17:00:56","doi":"10.21203/rs.3.rs-9039478/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":"3a3d5412-5cab-4596-a704-e4f4ba79662d","owner":[],"postedDate":"April 30th, 2026","published":true,"recentEditorialEvents":[{"type":"decision","content":"Rejected","date":"2026-05-16T14:07:30+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-16T11:51:17+00:00","index":27,"fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-05-04T12:46:42+00:00","index":25,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-16T14:24:52+00:00","versionOfRecord":[],"versionCreatedAt":"2026-04-30 17:00:56","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9039478","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9039478","identity":"rs-9039478","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}