一种虚拟场景车辆检测与测距方法

Vehicle Detection and Ranging in Virtual Scene

为了提升前方车辆测距精度,提出了一种基于改进目标检测(SSD)算法的车辆检测与测距方法。采用场景仿真软件PreScan搭建测试场景,使用改进的SSD算法,对摄像头采集的视频进行车辆目标检测。将检测框下边缘的中心点坐标作为图像中车辆坐标,根据单目视觉测距算法进行车辆目标测距。实验结果表明:在网络阈值为0.9时,改进的SSD算法对小车辆目标检测精度提升了6.6%。可以检测到70 m左右的目标,提出的测距方法可以达到2%的精度。

In order to improve the accuracy of vehicle ranging,a vehicle detection and ranging method based on improved single shot multibox detector(SSD)was proposed.Prescan software was used to build the test scene,and the improved SSD algorithm was used to detect the vehicle target in the video captured by the camera.The coordinates of the center point of the edge under the detection frame were taken as the vehicle coordinates in the image.The vehicle target was located according to the monocular vision ranging algorithm.The experimental results show that when the network threshold is 0.9,the improved SSD algorithm increases the detection accuracy of small vehicle target by 6.6%.The improved SSD algorithm detects the target of about 70 m,and the proposed ranging method achieves 2%accuracy.