一种改进的基于平面先验的多视图密集匹配算法OA

To address the issue of reduced model accuracy caused by the assumption of non-planar regions as planar regions during the construc-tion of a planar prior using the ACMP algorithm,we put forward an improved multi-view dense matching algorithm based on planar prior.Firstly,we used Canny operator to detect image edges,and used the detection results to remove non-weak texture regions.Then,we used the images that removed non-weak texture regions to construct a weak texture region planar prior model,and used this model to implement the ACMP algorithm based on planar prior.In the subsequent sampling calculation process of depth map fusion,we used the weak texture region planar prior model to oversample on weak texture regions and perform a small amount of sampling on non-weak texture regions,which could reduce the density of point clouds in planar regions while preserve more point clouds details in non-planar regions.Experimental results show that the number of point clouds in the planar region of experimental data has decreased by about 43.1%,whereas the number in the non-planar region has increased by about 65.5%.Compared with existing methods,the improved algorithm enriches the model details of non-weak texture regions and improves the integrity of point clouds.