双目立体视觉的目标识别与定位  

Target recognition and location based on binocular stereo vision

作　　者：尚倩[1,2] 阮秋琦[1,2] 李小利[1,2] 

机构地区：[1]北京交通大学信息科学研究所,北京100044 [2]北京交通大学现代信息科学与网络技术北京市重点实验室,北京100044

出　　处：《智能系统学报》2011年第4期303-311,共9页CAAI Transactions on Intelligent Systems

基　　金：国家自然科学基金资助项目(60973060);高等学校博士学科点专项科研基金资助项目(20080004001)

摘　　要：双目立体视觉系统可以实现对目标的识别与定位.此系统包含摄像机标定、图像分割、立体匹配和三维测距4个模块.在摄像机标定部分,提出了基于云台转角的外参数估计方法.该方法可以精确完成摄像头旋转情况下外参的估计,增强了机器人的视觉功能.并利用广茂达机器人系统,基于改进的双目视觉系统进行目标识别与定位,以此结果作为依据控制机器人的手臂进行相应运动,最终实现了对目标物体的抓取,验证了提出方法的可行性.A binocular stereo vision system can be used for target recognition and location. This system consists of four modules, namely camera calibration, image segment, stereo matching, and binocular distance, furthermore, in the calibration section, a method of estimating the outer parameters based on the head angle was studied. In the experiment, this method could estimate the outer parameters accurately when the camera rotated, enhancing the vi- sion function of the system. Then this improved binocular system was realized by the Grandar Robot platform. It was used for target recognition and location, and the results were used to control the robot motion. In the end, the target could be captured successfully, so the proposed method was proven to be feasible.

关 键 词：双目立体视觉 摄像机标定 图像分割 立体匹配 三维测距 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]