斜面参数优化全局立体匹配算法研究

针对现有立体匹配算法在非平行平面区域匹配中出现“阶梯效应”的问题，提出一种斜面参数优化的全局立体匹配算法。该算法用斜面参数替代视差值作为全局匹配算法的优化变量，并结合粒子滤波思想实现斜面参数的（近似）连续取值及能量函数的连续域全局优化，理论上可以同时消除传统匹配算法中视差离散取值及视差一致性约束带来的影响，从根本上消除“阶梯效应”。针对典型非平行平面测试图像对的实验结果表明，该算法在有效消除“阶梯效应”的同时降低了误匹配率。

Stereo matching based on global optimization of slant-plane

A novel stereo matching algorithm based on global optimization of slant-plane is proposed in this paper on the purpose of eliminating the “stair-case” problem appearing in the regions of non-parallel plane. Firstly, slant-plane replaces disparity as the optimization parameter in the global energy function to make wiser constraint for non-parallel planes. Secondly, the idea of particle filter is used to get an approximate continuous disparity while the new energy function is optimized. In theory, the proposed algorithm can eliminate the “stair-case” fundamentally in the reason of slant-plane consistency constraints and continuous disparity. Experiments on typical slant-plane scene indicate that the proposed algorithm can eliminate “stair-case” effectively as while as increasing matching rate.