基于视差图像的重叠果实图像分割算法  

Image Segmentation Algorithm for Overlapping Fruits Based on Disparity Map

作　　者：彭辉[1] 吴鹏飞[1] 翟瑞芳[2] 刘善梅[1] 吴兰兰[3] 景秀[1] 

机构地区：[1]华中农业大学理学院,武汉430070 [2]华中农业大学计算机应用研究所,武汉430070 [3]华中农业大学工学院,武汉430070

出　　处：《农业机械学报》2012年第6期167-173,共7页Transactions of the Chinese Society for Agricultural Machinery

基　　金：国家自然科学基金资助项目(41101409);中央高校基本科研业务费专项资金资助项目(2011JC020)

摘　　要：为解决自动采摘视觉系统中重叠果实的分割问题,提出了基于视差图像的果实分割算法。采用双目立体视觉系统获取图像对,对图像对进行预处理和校正,通过图像对的立体匹配来获取视差图像,最后对视差图像进行分割。该算法将分割的依据和信息从二维图像的颜色、形状、纹理等扩展到三维空间的深度,对空间距离不同的目标具有较好的分割效果。实验表明,对获取的视差图像进行基于区域的分割时,其区域间灰度对比度为0.98,目标计数一致性达到0.90;进行基于边缘的分割时,其边缘检测误差为5.74%,因此,该方法对重叠果实区域的分割是有效的。To solve the problem of segmentation for overlapping fruits, an image segmentation algorithm based on disparity map was developed. Firstly, binocular stereo images were obtained by binocular stereo vision system. Then, these images were preprocessed and rectified. Thirdly, stereo matching for rectified images pair to get disparity values for every pixel was made. At last, the disparity map was generated. Distance information in 3-D space was added to algorithm besides color, shape and texture information in 2-D space, so the fruits with different distances in actual space were segmented better in disparity map than in normal image. Experimental results showed that the accuracy of segmentation for disparity map by area-based method was 0.90, and the edge detection error was 5.74%. The proposed method was valid for segmentation of overlapping fruits.

关 键 词：重叠果实 双目立体视觉 视差图像 图像分割 

分 类 号：S126[农业科学—农业基础科学] TP391.41[自动化与计算机技术—计算机应用技术]