基于YOLOv3的雾天道路目标检测  

Fog Road Target Detection Based on YOLOv3

作　　者：蒲虹林 田怀文[1] 乐思显 PU Honglin;TIAN Huaiwen;LE Sixian(School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610000)

机构地区：[1]西南交通大学机械工程学院,成都610000

出　　处：《计算机与数字工程》2023年第9期2119-2124,共6页Computer & Digital Engineering

摘　　要：目前道路目标检测算法研究多基于正常天气,对雾天场景下道路目标检测研究较少,故提出一种基于YOLOv3的雾天道路目标检测算法。为提高检测设备在恶劣天气下对道路目标的检测能力和检测速度,利用改进后的金字塔池化结构构建去雾模块,并将其嵌入至YOLOv3目标检测网络中;引入通道注意力机制,提高DarkNet53的图像特征信息提取能力;增加一层检测层提高小目标物体检测能力,实现了一种去雾网络和检测网络联合优化的道路目标检测网络A-YOLOv3。利用大气散射模型和图像景深信息自制雾天道路数据集S-KITTI并将其用于实验验证,实验结果表明:通过改进优化使得模型检测精度从65.22%提升至72.5%。At present,the research on road target detection algorithm is mostly based on normal weather,and there is little research on road target detection in foggy scenes,so a foggy road target detection algorithm based on YOLOv3 is proposed.In order to improve the detection ability and detection speed of road targets in bad weather,the improved pyramid pooling structure is used to construct a defogging module and embedded in the YOLOv3 target detection network.The channel attention mechanism is introduced to improve the image feature information extraction ability of DarkNet53.By adding a detection layer to improve the detection ability of small target objects,a road target detection network A-YOLOv3 jointly optimized by defogging network and detection network is realized.The atmospheric scattering model and image depth of field information are used to make the foggy road dataset S-KITTI and then uses it for experimental verification,and the experimental results show that the detection accuracy of the model is improved from 65.22% to 72.5% through improved optimization.

关 键 词：YOLOv3 金字塔池化 雾天 联合优化 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]