Temporal Tampering Detection in Automotive Dashcam Videos Using Frame Difference Features and a 1D Convolutional Neural Network

Automotive dashboard cameras are widely used to record driving events and often serve as critical evidence in accident investigations and insurance claims. However, the accessibility of free and low-cost editing tools has increased the risk of video tampering, emphasizing the need for reliable methods to verify video authenticity. Temporal tampering typically involves manipulating the order of frames through insertion, deletion, or duplication. This study proposes a deep learning approach for detecting such tampering in dashcam videos using frame-difference analysis. Frame differences are calculated as the absolute pixel-wise difference between consecutive frames, and the resulting values are aggregated into a temporal magnitude signal, which is then normalized to the range [0, 1]. Sudden spikes in this signal indicate potential tampering. These normalized sequences are processed by a one dimensional Convolutional Neural Network (1D-CNN) for classification. Experimental evaluation using a custom dataset derived from the D2-City dataset demonstrates strong performance, achieving detection accuracies of 98.9% for frame insertion, 94.4% for frame deletion, and 92.4% for frame duplication. The model was further extended for multi-class classification to distinguish among non-tampered, insertion, deletion, and du-plication videos, achieving 92.8% accuracy with consistent precision, recall, and F1-scores. These results, combined with low computational cost, indicate strong potential for near real-time deployment in automotive cybersecurity and forensic applications, enhancing trust in IoT-enabled transportation systems.