雨天车辆检测的两阶段渐进式图像去雨算法

Two-Stage Progressive Image Deraining Algorithm for Vehicle Detection in Rainy Days

为了提升雨天车辆检测的精度,解决智能网联汽车的车辆检测系统受雨纹干扰导致精度衰退的问题,提出一种雨天车辆检测的两阶段渐进式图像去雨算法。该算法搭建了以轻量级特征提取与加权模块、高效率特征传递与融合模块为核心的两阶段渐进式去雨网络,实现了对雨纹信息的挖掘与捕获,完成了雨纹的精准去除。为了验证所提算法的有效性,融入基准车辆检测器YOLOv5,对输入YOLOv5的去雨图像进行检测。同时根据智能网联汽车的工作环境构建了混合车辆数据集。在该数据集上的结果表明:雨天交通场景下,相比其他算法,所提去雨算法对基准车辆检测器YOLOv5的精确率、召回率、mAP@0.5的增益分别为3.0个百分点、8.9个百分点、7.6个百分点,证明所提去雨算法能够显著提升对雨天车辆的检测精度,可应用于实际场景。

This study proposes a two-stage progressive image deraining algorithm for vehicle detection in rainy days.The proposed algorithm aims to improve the accuracy of vehicle detection in rainy days and solve the problem of accuracy degradation caused by rain streak interference in the vehicle detection system of intelligent and connected vehicles.For the algorithm,a two-stage progressive deraining network was developed with light feature extraction and weighting block along with efficient feature transfer and fuse block as the core to realize the mining and capture of rain streak information and achieve accurate deraining.The deraining images were input to benchmark vehicle detector YOLOv5 for verifying the effectiveness of the proposed algorithm.Furthermore,a mix vehicle dataset was constructed based on the working environment of intelligent and connected vehicles.The gains of the proposed deraining algorithm on the precision,recall,and mAP@0.5 of the benchmark vehicle detector YOLOv5 are 3.0 percentage points,8.9 percentage points,and 7.6 percentage points,respectively,under a rainy traffic scenario compared with other algorithms.The results prove that the proposed deraining algorithm considerably improves the accuracy of vehicle detection in rainy days and can be used in practice.