基于快速伪球滤波的智能拖拉机视觉导航中场景去雾方法  

Dehazing Method Based on Fast Pseudosphere Filtering for Visual Navigation of Intelligent Tractor in Haze Weather Scene

作　　者：张征明[1] 卢伟[1,2] 陆静霞[1] 

机构地区：[1]南京农业大学工学院江苏省农业智能化装备重点实验室,江苏南京210031 [2]江苏省远程测控重点实验室,江苏南京210096

出　　处：《机器人》2015年第5期603-613,640,共12页Robot

基　　金：国家自然科学基金青年基金资助项目(51405239);江苏省自然科学基金青年基金资助项目(BK20130696);中央高校基本科研业务费专项资金资助项目(KYZ201427);远程测控技术江苏省重点实验室开放基金资助项目(YCCK201501)

摘　　要：为适应智能拖拉机雾霾天气下道路的视觉导航,同时针对作业环境图像背景中天空居多的特征,提出一种基于快速伪球滤波的去雾方法.首先根据滤波器参数利用降秩逼近的方法对视觉图像产生滤波模板.而后利用伪球滤波保持边缘的平滑特性获得准确的暗通道图和大气透射图.最后,利用大气耗散模型的修正解快速恢复无雾图像.实验结果表明,快速伪球滤波方法与导向图滤波、Tarel中值滤波、多尺度Retinex、小波域Retinex方法相比,对图像去雾效果综合评价指标值分别提高54.7%、37.6%、35.2%、44%,算法耗时约0.18 s.能够满足智能拖拉机视觉导航实时性的要求.In order to make the visual navigation of intelligent tractors adapt to the foggy weather, an image dehazing method using the fast pseudosphere filter is proposed, to get a clear image from a foggy picture with the sky as main back-ground. Firstly, the filter template for the visual images is obtained based on filter parameters by using the reduced rank approximation method. Then, the edge smoothness is guaranteed through pseudosphere filtering to get the exact dark channel and the atmosphere scattering veil. Finally, a clear view of the image is recovered quickly with the correction solution of the atmospheric dissipation model. The experimental results show that this method can improve the objective assessment indica-tor of the image dehazing by 54.7%, 37.6%, 35.2%and 44%compared with the guided image filtering method, Tarel median filtering method, multi-scale retinex method and the wavelet-based retinex method. Furthermore, the time consumption of this algorithm is about 0.18 s, which can meet requirements of the intelligent tractors for real-time visual navigation.

关 键 词：图像去雾 伪球滤波 降秩逼近 智能拖拉机 视觉导航 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]