A three-dimensional reconstruction method based on point cloud optimization sampling

Abstract

The invention discloses a three-dimensional reconstruction method based on point cloud optimization sampling and belongs to the field of computer vision. The objective of the invention is to solve theproblem that in the three-dimensional reconstruction process of dense point cloud, under the condition that the main texture characteristics are ensured, the reconstruction speed is low. According tothe method, a mean shift algorithm is mainly utilized to carry out region segmentation on a plurality of corrected images; an improved bidirectional DP algorithm is adopted; sequentially carrying outstereo matching on the same regions after the adjacent images are segmented to obtain a disparity map; secondly, eliminating interference noise of the disparity map by using bilateral filtering to obtain a plurality of dense depth point cloud maps, then obtaining main texture features through self-adaptive random sampling, removing smooth areas, finally calculating depth value estimation of eachpoint cloud, and extracting the main point cloud according to the sequence of the estimation values. According to the method provided by the invention, more obvious texture features can be obtained ata higher calculation speed, and the method has a wide application prospect in natural scene three-dimensional reconstruction.

Description

technical field [0001] The present invention relates to the field of computer vision graphics, and more specifically, to a three-dimensional reconstruction method based on point cloud optimal sampling. Background technique [0002] In visual 3D reconstruction, stereo matching is the core link, which can be divided into local matching algorithm and global matching algorithm. Since the global matching algorithm uses the conditions of global constraints, the local area of ​​the image is not blurred, so the global stereo matching algorithm is commonly used in 3D reconstruction. The commonly used global stereo matching algorithms include graph cut (GC), belief propagation, and dynamic programming ( DP) and so on. The traditional dynamic programming algorithm uses the global energy minimum solution to divide a large decision-making problem into several sub-problems, so as to obtain the optimal disparity map. The stereo matching algorithm based on traditional dynamic programming ...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the deficiencies of the above-mentioned prior art, the present invention provides a 3D reconstruction method based on point cloud optimal sampling, the purpose of which is to solve the problem of large calculation cost and the disparity map obtained by matching in the traditional dynamic programming-based global stereo matching algorithm. There is a problem of stripes, and at the same time solve the problem of how to remove the smooth area to retain the main texture information in the reconstruction process, and how to retain only the accurate and necessary point cloud information in the multi-view fusion process

Images