Computer Vision Application in Management of Vehicle Speed in Factory Area

This paper presents a solution for managing vehicle speeds in specific areas such as factories or industrial facilities. Firstly, the videos, captured by fixed cameras, are processed to extract and detect vehicles by the fifth version You Only Look Once (YOLOv5). Then, the tracking-by-detection method, namely Strong Simple Online Real Time Tracking (StrongSORT), is used to track vehicle trajectories across consecutive frames by maintaining an identifier assigned to each vehicle. Next, these models' output results are obtained to estimate vehicle speeds. Two-speed estimation methods, namely Touch-two-frame and Distance-in-line, which we called C2V, KCKH respectively, in this study, are conducted and compared in various data contexts and parameters such as distance between two lines in the C2V and the number of calculated frames in KCKH method. Finally, over-speed vehicles are identified and sent to the Telegram application for the violation storage. The experimental results show that the speed measurement results in the C2V method yield higher accuracy when compared to the KCKH method. While the lowest RMSE of C2V is 0.93 (km/h), the lowest RMSE of KCKH is 4.83 (km/
$h$
). Furthermore, in the C2V method, when changing the calculated distance or the rotation angle of the camera, the obtained results do not differ too much.