三目自适应权值立体匹配和视差校准算法

立体匹配是计算机视觉研究中的关键问题.相比于双目视觉,三目视觉能够获得更多的信息和额外的极线约束消除立体匹配的歧义性.为了提高三目立体匹配的精度,提出一种基于自适应权值和视差校准的三日立体匹配方法.将双目视觉中有效的自适应权值窗选择算法应用到三目视觉中,进行匹配窗的选择;提出一种新的目标图像选择算法,能够合理利用平行基线三目立体视觉系统中不同目标图像提供的信息,有效地消除遮挡,提高匹配的精度;提出一种适用于三目视觉的视差校准算法,利用三目图像像素间的色彩相似性和距离约束将初始匹配的视差结果进行校准,得到最终的视差图.实验结果表明,本文算法结构简单,能够生成浓密、高精度的视差图.     

Robust view selection in multiview stereo

A robust view selection algorithm for multiview stereo matching is presented. Different from the existent view selection algorithms which pick only the believable views for matching, this method assigns an adaptive weight to each target view based on the dissimilarity between it and the reference view. So it can utilize the information of all views more sufficiently. The algorithm has been evaluated with different real images to demonstrate its robustness.