基于双目立体视觉的复杂背景下的牛体点云获取  

Acquisition of point clouds of cattle in complex situation based on binocular stereo vision

作　　者：薛广顺 来智勇[1] 张志毅[1] 王美丽[1] 

机构地区：[1]西北农林科技大学信息工程学院,陕西杨凌712100

出　　处：《计算机工程与设计》2015年第5期1390-1395,共6页Computer Engineering and Design

基　　金：陕西省农业科技创新基金项目(2012NKC01-20)

摘　　要：提出一种基于双目立体视觉的复杂背景下的牛体点云获取方法。在点云获取过程中,针对牛不是保持不动的实际情况,采用基于被动视觉技术的双目立体视觉方法;在相机标定过程中,进行单目标定获取相机内部参数,统一标定获取相机外部参数,针对牛体毛色及所处环境复杂的情况,采用基于贝叶斯的皮肤检测算法提取牛体图像,采用基于SIFT的特征点提取和匹配方法实现特征点的匹配,利用计算机视觉中的极线几何原理,剔除误匹配点,通过相机的成像模型求取牛体的三维点云。实验验证了该方法能够在复杂背景下获取牛体点云,并得到较好的点云数据,较好解决了双目立体视觉中相机标定和立体匹配两个较关键和困难的问题。To obtain the point cloud data of cattle in a complex situation,a method based on binocular stereo vision was presen-ted.Considering that the cattle do not always stand motionless during the process of obtaining point cloud,the binocular stereo vision of passive vision technology was adopted.During camera calibration,single camera calibration was conducted firstly,and then extrinsic parameters were obtained through calibrating the two cameras at the same time.To get the cattle from the com-plex situation,a skin detection algorithm based on Bayesian classification was implemented.Feature points extraction and matc-hing method based on SIFT algorithm was employed and epipolar geometric principle was utilized to eradicate the false matching points.Finally,the three-dimensional coordinate was gotten by means of imaging model of camera.The experimental results show that the proposed method is feasible in the complex situation,and it can obtain good point cloud data.The two critical and difficult issues of camera calibration and stereo matching in binocular stereo are solved availably.

关 键 词：双目立体视觉 相机标定 SIFT特征点 特征匹配 贝叶斯分类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]