基于偏振成像和YOLOv8的雾天道路目标检测  

Road Object Detection in Foggy Weather Based on Polarization Imaging and YOLOv8

作　　者：谈爱玲[1] 李晓航 赵勇[2] 高美静[3] 苏海杰 刘闯 郭天安 TAN Ailing;LI Xiaohang;ZHAO Yong;GAO Meijing;SU Haijie;LIU Chuang;GUO Tianan(The Key Laboratory for Special Fiber and Fiber Sensor of Hebei Province,School of Information and Science Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;The Key Laboratory of Measurement Technology and Instrumentation of Hebei Province,School of Electrical Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;College of Integrated Circuits and Electronics,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]燕山大学信息科学与工程学院河北省特种光纤与光纤传感重点实验室,河北秦皇岛066004 [2]燕山大学电气工程学院河北省测试计量技术及仪器重点实验室,河北秦皇岛066004 [3]北京理工大学集成电路与电子学院,北京100081

出　　处：《计量学报》2024年第11期1626-1633,共8页Acta Metrologica Sinica

基　　金：国家自然科学基金(61971373);河北省自然科学基金(F2023203077)。

摘　　要：雾天天气下,车辆和行人目标的准确检测对汽车自动驾驶非常重要。首先通过偏振成像装置采集了0°、45°、90°和135°角度的偏振图像,并通过3种不同的融合方式构建了I04590、stokes和pauli图像数据集。提出一种改进YOLOv8的目标检测算法以提高雾天偏振图像中汽车和行人两类目标的检测准确率。提出一种基于混合池化的MixSPPF结构,改善了原有SPPF结构对全局信息的提取能力;然后基于不同大小的卷积设计了Multiscale Module模块并结合Coordinate Attention注意力机制增强了对空间信息和通道信息的提取。实验结果表明,提出的改进YOLOv8算法获得的全类平均准确率P_(A)@0.5和P_(A)@0.5:0.95分别达到了83.4%和39.3%,比初始YOLOv8算法分别提升了1.6%和0.9%。Accurate detection of vehicle and pedestrian targets is crucial for autonomous driving in foggy weather.Detection of automobile and pedestrian targets in foggy weather is of interest to the field of autonomous driving.Polarized images at 0°,45°,90°,and 135°were first acquired by a polarization imaging device,then I04590,stokes,and pauli image datasets were constructed through three different fusion methods.An improved YOLOv8 object detection algorithm was proposed to improve the detection accuracy of two types of targets,automobiles and pedestrians,in polarized images in foggy weathers.A MixSPPF structure based on hybrid pooling was proposed to improve the original SPPF structure's ability to extract global information.Then a Multi-scale Module was designed based on convolutions of different sizes and combined with the Coordinate Attention mechanism to enhance the extraction of spatial and channel information.The experimental results showed that the proposed improved YOLOv8 algorithm achieved the mean average precision(mAP)is mAP@0.5 value of 83.4%and mAP@0.50.95 value of 39.3%,which were improved by 1.6%and 0.9%respectively compared to the original YOLOv8 algorithm.

关 键 词：目标检测 偏振图像融合 雾天天气 YOLOv8 MixSPPF Multi-scale Module 

分 类 号：TB96[机械工程—光学工程]