Violation Detection Algorithm for Trucks Right-turn Without Stopping Based on Video

With the increasing demand of the logistics industry and urban infrastructure, the activity of trucks within cities has increased, and accidents between right-turning trucks and vulnerable road users (VRUs) have received attention. In some regions, it is considered illegal for trucks to turn right without stopping. However, only manual visual inspection by the traffic management department is not enough to monitor such a large traffic volume. This paper proposes an automatic violation detection algorithm for right-turn non-stop trucks. Custom-trained YOLOv5 and DeepSORT are used to accomplish classification detection and real-time tracking of trucks. On this basis, a targeted illegal feature judgment module is designed for different road conditions in the real world to determine the illegal status of each truck. In the end, the algorithm outputs a fixed alarm message for illegal trucks. In real-life scenarios, YOLOv5 obtained an average accuracy (mAP) of 82% for truck detection, and the system could detect all violations in six video clips with a zero false alarm rate.However, there is no automatic detection algorithm for this violation. Therefore, it shows that the proposed detection algorithm for truck violations has good adaptability to truck right-turning violations under different monitoring images, and meets the current needs of traffic management departments.