一种基于PatchMatch的多视影像密集匹配改进方法

An Improved Method of Multi-view Image Matching Based on PatchMatch

多视密集匹配是倾斜摄影测量中的研究热点,提出一种基于改进的PatchMatch多视密集匹配算法。算法采用自适应归一化互相关作为匹配代价,由PatchMatch算法迭代运算优化得到影像深度图,再基于可见性融合深度图得到场景的三维点云。该方法在匹配所需影像的选择上,采用了同时考虑视图层和像素层的影像选择方法,在测度选择上,提出了具有边缘特性的相似性测度。最后,对影像进行密集匹配试验,结果表明,产生的点云质量得到一定的改善,验证了该方法的有效性。

Dense matching with Multi-view is a research hotspot in tilting photogrammetry. This paper proposes a multi-view matching algorithm based on improved PatchMatch. The algorithm uses the adaptive normalized cross correlation as the matching cost. The image depth map is obtained by the iterative operation of the PatchMatch algorithm, and then the 3 D point cloud of the scene is obtained based on the visible fusion depth map. This method adopts a method of image selection considering both the view layer and the pixel layer, and proposes a similarity measure with edge characteristic in the selection of the matching image. Finally, the dense matching experiment of the image shows that the quality of the generated cloud is improved and the effectiveness of the method is verified.