作者:陈秀锋，王成鑫，吴阅晨，谷可鑫
Authors:CHEN Xiufeng,WANG Chengxin,WU Yuechen,GU Kexin摘要:针对城市道路车辆检测中小目标车辆漏检率高和存在异类冗余框的问题，提出一种改进 YOLOv5s 的车辆实时检测算法 。对 YOLOv5s 算法网络结构进行优化，采用增加小目标检测层，将浅层特征图与深层特征图拼接后进行检测的方法，提升小目标车辆的检测率; 针对异类冗余框问题，采用加权非极大值抑制融合两边框信息的方法，提升检测准确性 。实验结果表明，改进 YOLOv5s 算法的平均检测精度( mAP@ 0. 5 ∶ 0. 95) 达到 64. 17% ，相 比 YOLOv5s 算法，查准率、召回率分别提高 1. 72% 、0. 72% ; 在小目标车辆检测中，正检率提高 5. 95% ，漏检率降低 4. 63% 。 改进YOLOv5s 算法能有效改善小目标车辆的检测精度和准确率。
Abstract:Aiming at the high missed detection rate of small target vehicles and the heterogeneous redundant frames in video vehicle detection，a real-time vehicle detection algorithm based on improved YOLOv5s was proposed. To improve the detection rate of small target vehicles，an optimization of the YOLOv5s algorithm network structure was established，which added a small target detection layer and spliced the shallow feature map with the deep feature map in the detection. For the problem of heterogeneous redundant frames，weighted non-maximum value suppression is used to fuse the information of both frames to improve the detection accuracy. The experimental results show that the average detection accuracy ( mAP @ 0. 5 ∶ 0. 95 ) of the improved YOLOv5s algorithm reaches 64. 17% . Compared with the YOLOv5s algorithm，the precision and recall rate are improved by 1. 72% and 0. 72% respectively. In the small target vehicle detection，the positive detection rate is increased by 5. 95% and the missed detection rate is reduced by 4. 63% . The improved YOLOv5s algorithm can effectively improve the detection precision and accuracy of small target vehicles.

PDF全文下载地址:

可免费Download/下载PDF全文