摘要
在车辆检测过程中,由于复杂背景影响较大、远场景车辆目标及遮挡目标特征难以提取,导致出现错检、漏检现象。因此,提出一种基于YOLOv5所改进的复杂背景下的车辆检测算法YOLOv5-BA。通过引入加权双向特征金字塔网络(BiFPN)特征融合思想,并在检测部分引入自适应特征融合模块ASFF来提高检测性能。实验结果表明,改进算法在KITTI数据集上检测精度达到了98%,检测速度FPS达到了42,检测精度高于目前主流的目标检测算法。在满足实时性检测要求的前提下,在复杂背景、远场景以及遮挡情况下,YOLOv5-BA具有更优异的表现。
During the process of vehicle detection,due to the great influence of the complex background,as well as the difficulty in extracting the features of vehicle targets and occluded targets in the distant scene,false detection and missed detection occur.Therefore,an improved vehicle detection algorithm YOLOv5-BA under the complex background of YOLOv5 is proposed.By referring to the feature fusion idea of Weighted Bidirectional Feature Pyramid Network(BiFPN),and introducing the adaptive feature fusion module ASFF in the detection part,the detection performance is improved.The experiment results show that the improved algorithm has a detection average precision of 98%on the KITTI dataset,and a detection speed of 42 FPS,and the detection accuracy is higher than the mainstream target detection algorithm.Under the requirements of real-time detection,YOLOv5-BA has better performance in complex background,far scene and occlusion situations.
作者
朱宝全
马长旺
宋亚伟
唐佳乐
赵强
ZHU Bao-quan;MA Chang-wang;SONG Ya-wei;TANG Jia-le;ZHAO Qiang(School of Communications,Northeast Forestry University,Harbin 150006,China;Scene Driving Test and Experience Division,SAIC-GM-Wuling Automobile,Liuzhou 545000,Guangxi Zhuang Autonomous Region,China)
出处
《信息技术》
2024年第6期54-60,66,共8页
Information Technology
基金
国家重点研发计划重点专项(2017YFC0803901)
中央高校基本科研业务费专项资金项目(2572016CB18)。
