基于双目视觉的基准差梯度立体匹配法??

因灰度相关只是从一个侧面来描述左右图像特征点区域之间的灰度相似性,没有考虑特征点之间的空间相关性,因此利用灰度间的相似性作为测量标准进行匹配,不可避免地出现误匹配,提出了在进行双目视觉立体匹配时,采用灰度相关匹配技术,提取复峰特征点作为初始匹配集,采用视差梯度有限约束优化初始匹配集.利用左右图像一对已知对应基准点,通过计算基准点与复峰集各点间的基准差梯度,采用基准差梯度极小化评判标准,确定唯一匹配,并将匹配结果确定为新的基准点以不断更新基准点,直至左(右)图像特征点匹配完毕.通过分别对一幅弱纹理实际自然图像及已知三维坐标标准件的三维重建,证实了所提方法的有效性和可靠性.

A Binocular Stereo Matching Algorithm Based on Reference Difference Gradien

Error matches will be happened inevitably by gray correlation alone due to gray correlation describing the gray similarity of feature points between the stereo image pairs too unilaterally, and their spatial correlations being not considered also. The gray correlation matching technique is utilized to extract multi-peak feature points as an original matching set while binocular stereo matching. Make use of gray gradient limit constraint to optimize the matching set. Adopt a pair of given corresponding reference points in the left and right images; and compute the reference gradients between the each point within the multi-peak set and the reference point. The unique match is obtained by minimal reference gradient. Take the got match as a new reference point and update new reference point continuously until all feature points in the left (or right) image being done. The algorithm mentioned is valid and reliable by 3D reconstruction on a pair of actual natural images with weak texture and the standard piece with given precise 3-D coordinates, respectively.