基于双目立体视觉的目标识别与定位

为从不同角度识别目标物体以及解决左右两幅图像中目标轮廓中心不匹配的问题, 将SURF(Speeded Up Robust Features)算法与GrabCut 算法相结合, 离线采集目标物体不同角度的图像, 生成目标模板图片库。利用SURF 算法完成目标物体的识别; 利用SURF 算法自动初始化GrabCut 算法, 实现目标轮廓的提取; 利用基于灰度相关的区域匹配算法完成目标轮廓中心点的匹配, 结合三维重建原理实现目标定位。实验结果表明, 该方法可以成功识别目标物体并对目标物体进行准确定位。

Object Recognition and Localization Based on Binocular Stereo Vision

In order to solve the issue that recognizing object from different angles and the mismatching of the contour center of the object in left and right images,this work combines the SURF( Speeded Up Robust Features)algorithm and GrabCut algorithm. Firstly,we capture object images from different angles to establish object template image database in offline mode and use SURF algorithm for object recognition. Secondly, automatically initialize the GrabCut algorithm using SURF algorithm and realize the object contour extraction. Finally,we realize the matching of the contour center of the object based on gray relevant regional matching algorithm and get the localization of the object according to 3d reconstruction principle. Experiments show that the proposed method can recognize object successfully and locate the object accurately.