Pedestrian Motion Estimation Using a Combination of Dense Optical Flow and Tracking by Detection
Ishika Soni, P. Vijayaraghavan, Anvesha Garg, Roshita David, Yogesh Kumar Sharma, T. Kuppuraj · 2024
Pedestrian motion estimation from video sequences has emerged as a crucial project for packages starting from surveillance to healthcare. To accurately estimate such movement calls for robustness in opposition to motion blur, occlusion, and illumination modifications. This paper proposes a novel estimation method that mixes dense optical glide and monitoring with the aid of detection to cope with these challenges. Dense optical go-with-the-flow offers sparse movement statistics, and the monitoring-by-means-of-detection method operates on the detected bounding packing containers. By means of combining both record sources, the proposed technique produces sturdy motion estimates of pedestrians within the video. Experiments on public datasets display the advantages of the mixture of optical waft and tracking-by-detection in terms of accuracy and robustness in opposition to motion blur, occlusion, and illumination adjustments.