基于深度学习的车辆检测与跟踪系统

Vehicle Detection and Tracking System Based on Deep Learning

通过对车流量信息的检测来分析各道路拥堵情况,交通部门可以及时对拥堵路段做出疏通响应,改善交通状况,降低交通事故率。车流量信息检测对于实时性和准确性要求都很高,分析设计了基于YOLOv3的车辆检测方法与Deep-SORT的多目标跟踪方法,满足了车流量检测系统的实时性和准确性。在GPU为NVIDIA 1080TI环境下,系统运行的帧率为21FPS,车流量的计数平均准确率为97.7%,该系统在实时性和准确性上都取得了较好的效果。

The detection of traffic flow information is very demanding for real-time and accuracy,so this paper analyzes and designs the vehicle detection method based on YOLOv3 and the multi-target tracking method of Deep-SORT,which satisfies the real-time and accuracy of the traffic flow detection system.When the GPU is NVIDIA 1080TI,the system runs at a frame rate of 21 FPS,and the average accuracy of the traffic flow count is 97.7%,which verifies that the system has achieved good results in real-time and accuracy.