A Video Axle Counting and Type Recognition Method Based on Improved YOLOv5S
Abstract: The number of axles of the vehicle and the type of tires can reflect the information of the vehicle to a certain extent, and the load capacity can be calculated according to the number of axles of the truck and the type of axles. Therefore, the identification of the axle is of great significance for judging whether the truck is overweight. At present, the method of calculating the axles is carried out by the method of Laser Radar or grating for axle counting. In the prior art, the method of axle counting is complicated to deploy and the cost is high. Some computer vision-based axle statistics methods have emerged in recent years, but complete vehicle sideways pictures are required. However, due to the long body of the truck and the limited space factor, it is difficult to obtain the complete vehicle in an original image. Although image stitching can solve this problem, the current video image stitching methods have a relatively high time cost. To solve this issue, we propose an object detection and tracking method based on YOLOv5s for axle counting and tire type identification. Experimental results show that the proposed method has extremely high accuracy and can meet real-time requirements even without GPU.
Loading