基于雾天模拟图像的能见度估计方法研究  

Visibility Estimation Based on Simulated Images of Foggy Weather

作　　者：邱实卓 叶青[1] 黄佳恒 刘建平 QIU Shizhuo;YE Qing;HUANG Jiaheng;LIU Jianping(School of Electrical and Information Engineering Changsha University of Science and Technology,Changsha 410000 China)

机构地区：[1]长沙理工大学电气与信息工程学院,长沙410000

出　　处：《电光与控制》2025年第3期94-100,共7页Electronics Optics & Control

基　　金：长沙理工大学研究生科研创新项目(CXCLY2022086)。

摘　　要：针对带有能见度标签的雾天图像数据集匮乏的问题,提出一种基于雾天模拟图像的能见度检测方法。通过无监督深度估计模型构建户外清晰图像的深度图,并且利用特征融合来细化深度图细节,采用暗通道法筛选天空区域估计大气光值,获得设定能见度下户外图像的透射率图,进一步获得带有不同能见度标签的雾天模拟图像数据集,基于此利用改进后的ShuffleNet V2网络训练能见度估计模型。对数据集和真实雾天图像进行了能见度等级估计的验证实验,实验结果表明,对能见度小于500 m的雾天图像具有良好的能见度估测结果,对小于200 m的检测准确率高于90%,总体准确率为87.8%,表明该方法具有可行性,可用于大雾条件下的道路能见度等级估测。Aiming at the shortage of fog image data set with visibility labels a visibility detection method based on simulation fog images is proposed.The depth map of clear outdoor images is constructed by unsupervised depth estimation model and the details of the depth map are refined by using feature fusion.The transmission map of outdoor images under set visibility is obtained by using dark channel method to estimate atmospheric light value and the simulation fog image dataset with different visibility labels is further obtained.Based on this the improved ShuffleNet V2 network is adopted to train the visibility estimation model.A verification experiment is conducted on the visibility grade estimation of the dataset and the real foggy images.The experimental results show that:1)The proposed method has good visibility estimation results for foggy images with visibility less than 500 meters;2)The detection accuracy is higher than 90%for foggy images with visibility less than 200 meters;and 3)The overall accuracy is 87.8%;which indicating that the method is feasible and can be applied to estimate the visibility level under fog conditions.

关 键 词：雾天模拟 深度图 特征融合 ShuffleNet V2 能见度估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]