基于机器视觉对交通场景中车辆速度识别研究

通过对YOLOv5机器视觉框架进行二次开发，同时融合DeepSORT追踪算法，实现对桥梁交通车辆时空信息的提取和车辆轨迹的追踪。改进了传统的虚拟线圈法，实现了对车辆速度的测量，避免对传统方法中因检测线圈的像素变化进行阈值的设定，提高了算法的普适性。最后，将算法应用到实际的场景中与测速仪结果进行对比，其中平均误差在1%以内，误差最大值控制在为15%以内。

Research on Vehicle Speed Recognition in Traffic Scene Based on Machine Vision

The extraction of spatio-temporal information of bridge traffic vehicles and the tracking of vehicle trajectories are realized by the secondary development of YOLOv5 machine vision framework, while incorporating DeepSORT tracking algorithm. Then the traditional virtual coil method is improved to achieve the measurement of vehicle speed, which avoids the setting of threshold value due to the pixel change of detection coil in the traditional method and improves the universality of the algorithm. Finally, the algorithm is applied to the actual scene to compare with the tachymeter results, in which the average error is within 1% and the maximum error is controlled within 15%.