SIGNATURE WORK
CONFERENCE & EXHIBITION 2023

Nowadays, online vehicle tracking and counting are important ways to manage traffic flow and mitigate traffic congestion. In this paper, classical online tracking methods including SORT and DeepSORT have been reviewed. Specific details and derivation steps have been compared. Furthermore, an experiment has been conducted with YOLOv8s pre-trained on the COCO dataset and the deep association metric pre-train on the VeRi dataset. DeepSORT Algorithm has been used for tracking. The four test videos are on the front-view cars, back-view cars, both directions, and crowed intersections respectively. The returned results include four video demos for visual presentation and tracking information tables for quantitative analysis. The MOTA results are satisfiable compared to the Multiple Object Tracking benchmarks. In the end, discussions on the ID switches problem, detection errors, undesired objects, and occlusion problems have been conducted by analyzing the video result demos. The possible reasons causing those problems and future work directions have been raised.