Real Time Object Identification and Movement Tracking through YOLO/DeepSORT and Optical Flow Techniques

This research presents a computer vision framework for real-time object detection, multi-object tracking, and motion analysis using YOLOv9, DeepSORT, and optical flow techniques. The proposed system integrates YOLOv9 for object identification, DeepSORT for maintaining object identities across video frames, and three optical flow algorithms—Lucas–Kanade, Farneback, and Horn–Schunck—for estimating object motion. The framework is evaluated using DIRS24-v1, a self-curated campus traffic dataset containing 2,374 high-resolution images and 5,144 annotated objects across seven traffic categories. Experimental results reported in the manuscript show that YOLOv9 achieves an mAP@0.5 of approximately 0.83, compared with 0.662 for the YOLOv8-OBB baseline. The DeepSORT tracking component is evaluated on a 129-frame video sequence, while optical flow analysis provides motion descriptors including speed, direction, and acceleration. The research demonstrates the integration of object detection, identity-preserving tracking, and motion estimation for traffic monitoring and intelligent surveillance applications. The implementation is evaluated using CPU-based hardware, highlighting the potential for deployment on systems with limited computational resources. Future work includes improving performance for underrepresented object classes, evaluating tracking robustness on larger public benchmarks, and optimizing the system for GPU-based real-time applications.