基于GIF-Shearlet算法的新旧土边界线视觉导航技术研究  

Visual navigation technology using guided image filter and shearlet algorithm based on the boundary line of new and old soil

作　　者：陈益杉 卢伟[1,2] 王玲[1] 田光兆[1] CHEN Yi-shan LU Wei WANG Ling TIAN Guang-zhao(College of Engineering and Key Laboratory of Intelligent Agricultural Equipment in Jiangsu Province, Nanjing Agricultural University, Nanjing, Jiangsu 210031, China Jiangsu Province Key Laboratory of Remote Measurement and Control Technology, Nanjing, Jiangsu 210096, China)

机构地区：[1]南京农业大学工学院/南京农业大学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：《农业现代化研究》2017年第2期343-351,共9页Research of Agricultural Modernization

基　　金：国家自然科学基金青年基金(31401291);江苏省自然科学基金青年基金(BK20130696);中央高校基本科研业务费专项资金(KYZ201427)~~

摘　　要：针对遥操作拖拉机驾驶机器人旋耕作业时,工作环境中作物行多样化、光照不均的特点,提出一种基于导向滤波(Guided Image Filter)和剪切波变换(Shearlet Transform)的方法用于提取新旧土边界线以完成拖拉机视觉导航。首先,将图像快速转换到YCr Cb颜色空间,对灰度化的图像进行导向滤波,然后使用Shearlet-canny算子提取新旧土的边缘信息,最后经过Hough变换给出视觉导航线。结果表明,在YCr Cb颜色空间对图像进行灰度化处理与HSV、HIS、RGB颜色空间相比,效果最好,耗时最短,分别快94.0%、94.3%和25.4%;导向滤波处理图像的方法与Tarel中值滤波、MRetinex滤波、小波域Retinex滤波及同态滤波相比,算法耗时分别短87.5%、79.5%、88.8%和87.0%;采用Shearlet-Canny算子检测边缘并经过Hough变换后提取的导航线精确度最高,最大误差小于0.5°。研究表明,基于导向滤波和剪切波变换的新旧土边界线提取方法用于拖拉机智能导航是可行的。In this paper, a leading line extracted method based on the boundary of new and old soil was proposed according to the characteristics of working environment such as diversification of crop rows and uneven illumination. Under rotary tillage mode, teleoperating tractor can work efficiently in real time with the help of the proposing leading line extracted method. Firstly, the gray image was rapidly transformed into YCrCb color model and filtered by guided image filter, then the edge information was extracted by using shearlet-canny operator. In addition, the leading line was obtained by Hough Transform. Finally, the experiments were carried out which showed that the gray processing under YCrCb color model was 94.0%, 94.3% and 25.4% faster than HSV, HIS and RGB respectively. Especially, with the help of guided image filter, the processing was 87.5%, 79.5%, 88.8% and 87.0% faster than Tarel filter, MRetinex filter, wavelet domain Retinex filter and homomorphic filter respectively. Moreover, shearlet-canny operator was the most efficient one which can get the most precise line whose deviation was less than 0.5°. In general, the method based on guided image filter and shearlet transform is available, it can help the smart tractor to meet the visual navigation needs and has an important application value.

关 键 词：智能拖拉机 遥操作 视觉导航 新旧土 导向滤波 剪切波变换 

分 类 号：S222.3[农业科学—农业机械化工程]