Research Information System
5th International Conference on Intelligence-Based Transformations of Technology and Business Trends, ICITTBT 2025, Tirana, Albania, 29 - 30 May 2025, vol.2669 CCIS, pp.708-726, (Full Text)
Real-time vehicle tracking plays an important role in intelligent transportation systems, traffic monitoring and autonomous driving technologies. This study presents the development of a real-time vehicle tracking system that combines the You Only Look Once (YOLO) deep learning-based object detection architecture with a custom stereo vision setup. The system is designed to detect, track and estimate the distance of moving vehicles in traffic scenes. Vehicles are detected with high accuracy using the YOLOv8 algorithm, and bounding boxes are generated for each detected object. For depth estimation, the disparity map is generated using the Semi-Global Block Matching (StereoSGBM) algorithm, which produces a dense and accurate depth map by evaluating pixel correspondences between stereo images. The resulting depth data is refined using a Kalman filter to suppress measurement noise and stabilize the trajectory of each tracked object. In addition, the direction and speed of motion of each vehicle is estimated using the Lucas-Kanade optical flow algorithm applied to key points within the bounding boxes. The fusion of depth and motion information allows the location of each object to be projected into real-world coordinates, visualized by a top-down map. This visualization provides an intuitive view of vehicle movement, allowing real-time analysis of traffic dynamics such as approach and reversing behavior. The proposed system provides a flexible, scalable and cost-effective solution for intelligent transport and traffic monitoring applications, with high potential for integration into decision support systems and smart city infrastructures.