基于改进YOLOv4的自动驾驶场景车辆检测

Vehicle Detection in Autonomous Driving Scenarios Based on Improved YOLOv4

针对自动驾驶场景,现有车辆检测算法对小目标车辆检测效果不好,导致车辆检测精度不高的问题,该文提出改进YOLOv4算法。首先增加小目标检测层,降低小目标车辆的漏检率;然后使用EIoU(efficient intersection over union)损失函数替换CIoU(complete intersection over union)损失函数,降低算法的边界框回归损失,提高算法的检测精度。在数据预处理阶段采用Mosaic数据增强的方法提高小目标车辆的训练效果,以及使用K-Means聚类算法选出更合适的检测锚框。在KITTI数据集上实验,改进算法平均检测精度为95.84%,检测速度为37.12帧/s,相比YOLOv4算法,平均检测精度提高2.84%。实验结果表明,改进YOLOv4算法达到了提高车辆检测效果的目的。

For autonomous driving scenarios,the existing vehicle detection algorithm has poor detection effect on small target vehicles,resulting in the problem of low vehicle detection accuracy. The improved YOLOv4 algorithm was proposed. Firstly,a small target detection layer is added to reduce the missed detection rate of small target vehicles.Then the EIo U(efficient intersection over union) loss function is used to replace the CIoU(complete intersection over union) loss function,which reduces the bounding box regression loss of the algorithm and improves the detection accuracy of the algorithm. In the data preprocessing stage,the Mosaic data augmentation method is used to improve the training effect of small target vehicles,and the K-Means clustering algorithm is used to select more appropriate detection anchor box. Experiments on KITTI dataset show that the mean average precision of the improved algorithm is95.84% and the detection speed is 37.12 frames/s. Compared with the YOLOv4 algorithm,the mean average precision is improved by 2.84%. The experimental results show that the improved YOLOv4 algorithm achieves the purpose of improving the vehicle detection effect.