复杂交通场景下自动驾驶道路目标检测

Road target detection for autonomous driving in complex traffic scenes

针对复杂交通场景下自动驾驶汽车对遮挡目标以及小目标漏检情况严重、精度较低的问题,本文提出改进的YOLOv4目标检测方法。首先提出一种新的非极大值抑制算法Soft-DIoU-NMS以提升算法对遮挡目标的精确定位能力,其次改进K-means聚类算法生成更准确的候选框,最后引入焦点损失缓解样本之间的不均衡问题。实验结果表明,改进后的YOLOv4检测精度达到89.91%,检测速度达到35.52 f/s,能够很好地解决复杂交通场景的目标检测问题。

Aiming at the problems of low accuracy and serious missed detection of occlusion targets and small targets by autonomous vehicles in complex traffic scenarios,an improved YOLOv4 target detection method is proposed in this paper.Firstly,a new non-maximum suppression algorithm—Soft-DIoU-NMS,is proposed to improve accuracy of the occlusion target location.Secondly,the K-means clustering algorithm is improved to generate a more accurate candidate box.Finally,the focus loss is introduced to alleviate the imbalance between samples.The experimental results show that the improved YOLOv4 detection accuracy reaches 89.91%and the detection speed reaches 35.52 f/s,which helps to solve the problem of target detection in complex traffic scenes.