基于立体匹配的低纹理图像重构算法

针对立体匹配中低纹理区域容易产生误匹配及传统动态规划固有的条纹问题，提出一种改进的基于双目立体视觉的低纹理图像三维重构算法。该算法首先基于像素间相似度和像素自身特异性计算匹配代价并引入一种自适应多边形支撑区域聚集匹配度。然后采用一种全局意义的简单树形动态规划进行逐点匹配。最后基于左右一致性准则运用一种简单有效的视差校正方法消除误匹配得到最终视差图。实验证明将算法运用于实拍低纹理灰度图像的匹配，得到轮廓光滑清晰的三维点云，说明该方法的适用性。

A Stereo Matching Based Construction Algorithm for Low Texture Images

Aiming at the mismatching problem in low texture areas and the well-known streaking effect of dynamic programming, an improved algorithm based on binocular stereo matching technology is proposed to generate three-dimensional reconstruction model for low texture images. Firstly, matching cost is computed based on the distinctiveness of pixels and the similarity among them. Secondly, an adaptive polygon-based support window is adopted in the matching cost aggregation, and a simple tree structure dynamic programming is introduced to guide the pixel to pixel matching. Finally, a simple and efficient method is presented to refine the mismatching pixels detected according to the left-right consistency constraint. To testify the applicability of the proposed algorithm, it is applied to low texture gray images captured in the real situation, and the experimental results show that smooth and vivid 3D points cloud models are generated.