基于参照物法的高速公路车距检测算法研究

Research on expressway distance detection algorithm based on reference object method

行车车距较近是导致高速公路发生交通事故的主要原因之一。为降低交管部门对高速公路未保持安全车距的违法行为的监管成本,基于高速公路场景下的卡口相机视频,提出一种基于参照物法的单目视觉高速公路车距检测方法。该方法首先使用YOLOv5+DeepSORT对视频中的多个车辆进行动态跟踪定位,从而获取车辆的像素位置,再选取长度已知的车道线作为参照物,通过拟合进行世界坐标系与像素坐标系转换,计算得到每个像素点位代表的实际距离,进而基于像素位置计算同一帧下的相邻车辆的实际距离。实验结果表明,在有效测距范围内,所提方法的车辆检测平均精度大于98%,测距误差小于6%,平均误差小于3%,所提方法的精确度能够满足交管部门的监管需求。

In the expressway,the close proximity between vehicles is a significant cause of traffic accidents.In order to reduce the cost of traffic control departments in supervising the illegal act of failing to keep a safe distance between vehicles on the highway,a monocular vision highway distance detection method is proposed based on the gantry camera video in the highway scene.The proposed method employs YOLOv5+DeepSORT to dynamically track and locate multiple vehicles in the video,enabling the acquisition of the pixel positions of vehicles.Subsequently,reference objects in the form of lane lines with known lengths are selected,and the world coordinate system and pixel coordinate system are transformed by fitting to calculate the actual distance represented by each pixel point.Finally,the actual distance between adjacent vehicles in the same frame is computed based on their pixel positions.Experimental results demonstrate that the proposed method achieves a vehicle detection mAP of over 98%,a ranging error of less than 6%,and an average error less than 3%,which meets the regulatory requirements of traffic control departments.