基于YOLOv7算法的车辆与行人检测  

Enhanced YOLOv7-based algorithm for vehicle and pedestrian detection

作　　者：吕天宝 臧景峰[1] 刘双林 贾庆阳 LYU Tian-bao;ZANG Jing-feng;LIU Shuang-lin;JIA Qing-yang(School of Electronic Information Engineering,Changchun University of Science and Technology,Changchun 130022,China)

机构地区：[1]长春理工大学电子信息工程学院,吉林长春130022

出　　处：《计算机工程与设计》2025年第3期903-909,共7页Computer Engineering and Design

基　　金：吉林省科技厅社会发展领域基金项目(20220203031SF)。

摘　　要：为解决目标检测算法在自动驾驶领域检测精度不佳、实时性差、适应性不足等问题,提出一种改进的YOLOv7算法。在算法网络中添加坐标卷积,加强算法对目标位置信息的捕获能力;增加通道和空间注意力机制CBAM到神经网络中,增强算法的特征捕获能力;使用EIoU作为算法的损失函数,增加损失收敛速度。在智能交通BDD100K数据集上进行分析实验,与基准算法YOLOv7对比,平均精度均值(mAP)增长了3.48%,达到了77.55%,各个类别的平均精度(AP)有所提高。实验结果表明,改进后算法能够有效实现不同环境下车辆与行人的目标检测。To solve the problems of low detection accuracy,poor real-time performance and insufficient adaptability of target detection algorithms in the field of automatic driving,an improved YOLOv7 algorithm was proposed.The coordinate convolution was added to the algorithm network to strengthen the ability of algorithm on capturing target location information.The channel and spatial attention mechanism CBAM was added to the neural network to enhance the feature extraction ability of the model.EIoU was used as the loss function of the algorithm,the loss convergence speed was accelerated.Analytical experiments were conducted on the Intelligent Transportation BDD100K dataset.The mean average precision(mAP) is increased by 3.48%,reaches 77.55%,compared to that of the benchmark algorithm YOLOv7.The average precision(AP) of each category is also improved.The experiments show that the improved algorithm can effectively achieve vehicle and pedestrian target detection in different environments.

关 键 词：车辆与行人检测 自动驾驶 深度学习 目标检测 坐标卷积 注意力机制 损失函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]