无人机导航场景中立体匹配算法研究

针对现有立体匹配算法难以兼顾无人机导航场景中匹配精度和幅度失真的鲁棒性,提出了一种基于改进Census变换的立体匹配算法。根据图像结构和色彩信息获得基于十字骨架的任意形状和大小的Census变换窗口,并利用改进Census变换的Hamming距作为匹配代价,采用局部优化（WTA）和左右一致性校验得到视差结果,获得高精度的视差图。实验表明,该算法不仅能够获得高精度的视差图,更对幅度失真有很强的鲁棒性,能够很好地适用于无人机导航场景。

Research on Stereo Matching in UAV Navigation Environment

Considering it is difficult for existing stereo matching algorithms to obtain high matching accuracy from images with amplitude distortion in UAV navigation,a novel algorithm based on Modified Census Transform is proposed.First,according to the image structure and color information,the arbitrary shaped window based on cross skeleton is constructed.Then,the matching cost based on Hamming distance is determined by Modified Census Transform,and the disparity and a high accurate disparity map can be obtained by Winner Take All and left-right consistency check.Experiments show that,the proposed algorithm produces high accurate disparity map and robust to amplitude distortion,which can adapt to UAV navigation environment.