基于Tao立体匹配框架的全局优化算法

Global method based on Tao stereo matching framework

传统的基于全局优化的立体匹配算法计算复杂度较高,在遮挡和视差不连续区域具有较差的匹配精度。提出了基于Tao立体匹配框架的全局优化算法。首先采用高效的局部算法获取初始匹配视差;然后对得到的视差值进行可信度检测,利用可信像素点和视差平面假设使用具有鲁棒性的低复杂度算法修正不可信任像素视差值;最后改进置信度传播算法,使其能够自适应地停止收敛节点的消息传播,并对经修正的初始匹配进行优化,提高弱纹理区域匹配准确度。实验结果表明,文中算法有效地降低整体误匹配率,改善了视差不连续及遮挡区域的匹配精度;同时,降低了算法整体复杂度,兼顾了速度,具有一定的实用性。

Traditional stereo matching based on global optimization is of computational complex which is poor to get accuracy matching result for the pixels in occlusion and depth discontinuity region. An efficient method of stereo matching was proposed, which was based on Tao stereo matching framework. Firstly, the initial matching disparity was obtained by the enhanced local method. Then occlusion and mismatched pixels were applied from reliable pixels using the robust method and named unreliable pixels, then reliable pixels and presupposition of disparity plane were used to refine the unreliable disparity. Finally, in order to improve the disparity accuracy in low texture region, an enhanced belief propagation method was used to optimize the refined initial disparity, which had adaptive convergence threshold. Experimental results demonstrate that our method can reduce the error matching rate effectively, improve the matching accuracy in occlusion and depth discontinuity region, reduce the computational complexity as well as improve matching speed.