基于YOLOv5s的行人与车辆检测算法研究
Research on improved pedestrian and vehicle detection algorithm based on YOLOv5s
朱立忠 1邵永斌 1杜海洋1
作者信息
- 1. 沈阳理工大学自动化与电气工程学院,辽宁沈阳 110159
- 折叠
摘要
针对城市交通环境复杂程度高,行人和车辆的检测结果精度偏低的问题.提出了改进的YOLOv5s行人与车辆检测算法.首先,在YOLOv5s加入SK注意力机制,同时选用GSConv模块替换网络中部分卷积模块,用于有效提升检测精度,同时保持网络参数量基本不变;其次,引入ECIOU损失函数,能够加快模型收敛;最后通过选用KITTI数据集来检验改进算法的效果.最终实验结果表明,改进后的YOLOv5s算法,在保证算法参数量基本不变的同时,可将行人与车辆平均检测精度从81.2%提升到了87.3%,验证了本次研究的有效性.
Abstract
Aiming at the problem of high complexity of urban traffic environment and low accuracy of pedestrian and vehicle de-tection results.An improved YOLOv5s pedestrian and vehicle detection algorithm is proposed.Firstly,the SK attention mechanism is added to YOLOv5s,and the GSConv module is selected to replace some convolutional modules in the network,which is used to effec-tively improve the detection accuracy while keeping the network parameters basically unchanged.Secondly,the ECIOU loss function is introduced,which can accelerate the model convergence.Finally,the KITTI dataset is selected to test the effect of the improved algo-rithm.The final experimental results show that the improved YOLOv5s algorithm can improve the average detection accuracy of pedes-trians and vehicles from 81.2%to 87.3%while ensuring that the number of parameters of the algorithm is basically unchanged,which verifies the effectiveness of this study.
关键词
YOLOv5s/行人车辆检测/注意力机制/损失函数优化Key words
YOLOv5s/Pedestrian and vehicle detection/Urban transportation/Optimization of loss function引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据