A Computer Vision based Vehicle Counting and Speed Detection System

The number of vehicles on the roads are increasing with every passing year. Appropriate measures are required to gain some information about the traffic density for traffic management. Moreover, higher the number of vehicles on roads, higher are the chances of rash driving and Overspeeding. This paper addresses the issue by proposing a vision-based approach to estimate vehicle speed and set up an overall vehicle counter as well a counter of vehicles belonging to different classes. This paper provides practical significance for traffic management on roads. The implementation requires three steps: video acquisition, object detection and multiple object tracking. After video acquisition, the task of vehicle detection is done using YOLOv5 which also classifies the vehicle. To track multiple vehicles in every passing frame of the video, we have used the StrongSORT algorithm which is an improvement of DeepSORT algorithm. The research experiment provided an accuracy of 85.27% for vehicle detection. The accuracy for speed of the vehicles was 87.9% with marginal room for errors from their ground truth values. Moreover, the model performs well in terms of counting the number of vehicles.