Research on Road Object Detection Algorithm Based on YOLOv5+Deepsort

In the Multiple Object Tracking (MOT) algorithm, usually adopt Sort based on Faster R-CNN as the target detector to identify and track the target. However, Sort hardly processes object occlusion, and the number of ID switches is very high the accuracy is very low in the case of object occlusion. In this paper, Deepsort is used to integrate apparent information and cascade matching to improve the recognition performance of Sort, so that the detection model can better deal with the case that the target is occluded for a long time and reduce the number of ID switches. At the same time, use YOLOv5 as the target detector, the performance of target detection is obviously better than Faster RCNN, and the consumption of GPU resources is less.