基于方向权值的立体匹配

Stereo matching using direction weight

立体匹配是三维重建问题的重要环节,良好的稠密视差图是恢复立体三维结构的先决条件。针对区域立体匹配策略、固定的窗口大小内视差不一致与匹配信息不够丰富的问题,提出了一种新的相似性测度函数和方向权值算法。首先,通过新的相似性测度函数建立两点的匹配代价;其次,在该点邻域内依据方向权值算法聚合窗口内的匹配代价;最后,通过WTA(Winner Take All)算法计算初始视差图,并根据左右一致性约束优化得到最终的稠密视差图。该方法较传统的区域匹配算法,匹配精度得到提高,算法运行效率也得到了提升,经实际验证该算法的稳健性较好。

Stereo matching is an important part of three-dimensional（ 3D） reconstruction problem and good recovery of dense disparity map is the precondition of 3D reconstruction. Based on regional matching strategy,a new similarity measure function was put forward,and the direction weight algorithm was applied to solve the problem between equal disparity and sufficient intensity variation within a fixed window size. Firstly,the matching costs between two points were constructed by the new similarity measure function. Secondly,the matching costs were aggregated in neighbourhood of the point. Finally,the initial disparity map was calculated by WTA（ Winner Take All） algorithm and optimized to achieve the final dense disparity map according to the left-right consistency constraint. This method has improved the matching accuracy and algorithm efficiency in comparison with traditional area-based algorithm. Experiment results show that the proposed method has good robustness.