基于改进YOLOV3-tiny目标算法的车辆测距技术研究

Application of improved YOLOV3-tiny target algorithm in vehicle ranging

在驾驶环境中有效的车辆测距利于驾驶员及时发现潜在危险,继而成为行车安全的重要保障。鉴于传统YOLOV3-tiny目标算法难以保证车辆测距的实时性与准确性。为此,通过修改YOLOV3-tiny网络的maxpool3层及新增upsample2层,实现了对传统YOLOV3-tiny目标算法的改进;与此同时,根据路道消失点检测算法构建动态逆透视模型,并通过对车辆目标方位进行判断完成车辆测距。结果显示,较之于Tiny YOLOv3算法,改进算法在平均精确率上提升了4.6%,在平均召回率上提升了7.4%;并且,车辆测距的动静态误差能够保持在7%以内,平均处理速度能够达到28帧每秒。结果表明,本研究改进的YOLOV3-tiny目标算法满足了车辆测距系统的准确性与实时性要求,继而成为行车安全的重要保障。

In the driving environment,effective vehicle distance measurement is conducive to the driver to discover the potential danger in time,and then become an important guarantee of driving safety.In view of the traditional YOLOV3-tiny target algorithm is difficult to ensure the real-time and accuracy of vehicle ranging.Therefore,by modifying maxpool 3 layer and adding upsample2 layer of YOLOV3-tiny network,the traditional YOLOV3-tiny target algorithm is improved;At the same time,according to the road vanishing point detection algorithm,the dynamic reverse perspective model is constructed,and the vehicle ranging is completed by judging the vehicle target orientation.The results show that compared with tiny yolv3 algorithm,the improved algorithm improves the average accuracy by 4.6%and the average recall by 7.4%;At the same time,in the application effect of vehicle ranging,not only the dynamic and static error can be kept within 7%,but also the average processing speed can reach 28 frames per second.In conclusion,the improved YOLOV3-tiny target algorithm meets the accuracy and real-time requirements of vehicle ranging system.