基于多策略融合的亚像素精度立体匹配研究

摘要

To reduce the mismatching in stereo matching at sub-pixel level and to improve matching accuracy,we propose a multi-strategy integration-based sub-pixel accuracy stereo matching method.It is mainly implemented in the following three steps:the processing of original image interpolation in combination with estimating the fractional disparity by minimum spanning tree-based cost aggregation strategy;providing new disparity search scope to determine the relations and decreasing the matching cost computation;introducing disparity plane in a large area of low-textured regions to fit and refining the disparity,smoothing on the accuracy at sub-pixel level,and reducing mismatching.Experiments show that the algorithm can effectively improve the matching accuracy to sub-pixel level while reducing mismatching.%为了减少亚像素立体匹配中存在的错误匹配以及提高匹配精度，提出基于多策略融合的亚像素精度立体匹配方法。通过以下三个步骤实现：原图像插值处理结合基于最小生成树代价聚集策略估计分数视差；给出新的视差搜索范围确定关系，减少匹配代价计算；在大片低纹理区域引入视差平面拟合细化视差，在亚像素精度上平滑，减少错误匹配。实验表明，算法有效地将匹配精度提高到亚像素级，同时减少了错误匹配。