基于改进Yolov4的行人测距技术研究

Research on Pedestrian Ranging Technology Based on Improved Yolov4

在自动驾驶技术的研究中,如何检测行人并测量其相对距离,辅助驾驶员预判风险一直是重点研究的问题。由于车辆行驶速度较快,行人又具有目标小、难检测等特征,使用原始的Yolov4模型搭配相似三角形测距法难以达到实时性检测和精准测距的要求。针对上述问题,通过修改Loss函数、轻量化网络结构等技巧改进Yolov4模型。实验结果表明,改进的Yolov4模型运行速度达到了平均62.5FPS,较Yolov4提升25%;结合改进相似三角形测距法,在纵向距离60米、横向距离4米内测距平均误差为4.90%,不仅运行速度更快而且测距准确性更高。

In the research of automatic driving technology,how to detect pedestrians and measure their relative distance,and how to assist drivers to predict the risk has always been the focus of research.Due to the fast speed of vehicles and the small and difficult detection of pedestrians,it is difficult to achieve the requirements of real-time detection and accurate ranging by using the original Yolov4 model and the basic similar triangle ranging method.To solve this problem,this paper improves the Yolov4 model by using the techniques of modifying the loss function and lightweight network structure and proposes an improved similar triangle algorithm based on the similar triangle algorithm for the spatial relationship between pedestrians and vehicles.The experimental results show that the running speed of the new Yolov4 model reaches 62.5 FPS on average,which is 25%higher than Yolov4.Combined with the improved similar triangle algorithm,the average distance measurement error is 4.90%in the longitudinal distance of 60 meters and the transverse distance of 4 meters,which not only runs faster,but also has higher ranging accuracy.