基于GCA-YOLOv5s的行人检测算法  

Pedestrian Detection Algorithm Based on GCA-YOLOv5s

作　　者：张求星 杨芳华 李峰 赵李萍 ZHANG Qiu-xing;YANG Fang-hua;Li Feng;ZHAO Li-ping(Institute of System Engineering,Academy of Military Sciences,Beijing 100166,China)

机构地区：[1]军事科学院系统工程研究院,北京100166

出　　处：《计算机仿真》2024年第11期199-204,共6页Computer Simulation

基　　金：国防科技基础加强计划技术领域基金(2020JCJQJJ362)。

摘　　要：针对智能网联汽车行人目标检测准确性和实时性较低等问题,提出了一种基于YOLOv5s改进的行人检测算法。首先,采用幻影模块替代传统卷积,在保证模型准确度的前提下,降低模型复杂度,从而提高模型实时性。然后,将坐标注意力模块引入特征提取网络获得重要特征,提升行人检测准确性。最后,针对损失函数计算的弊端改进边界框损失函数的计算方式,在现有损失函数中引入power变换,以获得更高的边界框回归精度。实验结果表明,使用改进模型在Widerperson数据集上进行实验mAP达到70.8%,相较原算法提升2.6%,检测速度达61FPS。所提算法较主流算法准确率和检测速度均有所提升。A pedestrian detection algorithm based on YOLOv5s improvement is proposed to address the issues of low accuracy and real-time performance in intelligent connected vehicle pedestrian target detection.First,the phantom module is used instead of the traditional convolution to reduce the model complexity while ensuring the model accuracy,thus improving the model real-time.Then,the coordinate attention module is introduced into the feature extraction network to obtain important features and improve the pedestrian detection accuracy.Finally,the calcu-lation of the bounding box loss function is improved to address the drawbacks of the loss function calculation,and the power transform is introduced into the existing loss function to obtain higher accuracy of the bounding box regression.The experimental results show that the experimental mAP using this improved model on the Widerperson dataset reaches 70.8%,which is a 2.6%improvement compared to the original algorithm,and the detection speed reaches 61 FPS.The proposed algorithm improves both accuracy and detection speed compared to the mainstream algorithm.

关 键 词：目标检测 坐标注意力 行人检测 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]