Three-dimensional reconstruction method based on fringe photograph collection of same scene

Abstract

A 3D reconstruction method based on scattered photo sets of the same scene is divided into three stages: the first stage: every pairwsie image feature matching and relative camera motion are estimated; and the stage is divided into 4 steps: (1) every two images are subject to the bidirectional nearest neighbor search and feature domain constraint to obtain a candidate correspondence; (2) the candidate correspondence is subject to parallax domain correspondence constraint to obtain a hypothesis correspondence; (3) the image coordinates of the hypothesis correspondence are standardized to solve an essential matrix estimation meeting the hypothesis correspondence; (4) the essential matrix is decomposed to obtain four groups of possible solutions of the camera motion, and the final solution is determined by the fault-tolerant forward depth constraint; the second stage: the optimized initial reconstruction camera pair is selected according to the results of the first stage, the standard sparse reconstruction method is applied, and the camera pose and the sparse geometric information of the scene are restored; the third stage: selective accurate and dense matching is carried out based on the results of second stage, and an accurate and dense 3D scene point cloud model is reconstructed by the triangulation method. The method has the advantages of obtaining reliable camera pose and high-density scene geometric information, greatly shortening the reconstruction time, having relatively high reconstruction efficiency, and is applicable to processing the scattered photo set with large data size.

Description

technical field [0001] The invention belongs to the field of virtual reality technology and computer vision. Specifically, three-dimensional reconstruction is carried out by scattered photos of the same scene, including restoring the shooting position, direction and field of view of the camera, and a dense 3D scene point cloud model. It is mainly used for Three-dimensional organization of digital photo resources on PC or Internet, as well as image-based drawing and browsing, etc. Background technique [0002] In computer vision, the 3D reconstruction theory and algorithm of multi-view geometry have been developed more maturely in the past two decades, document 1—R. Hartley, and A. Zisserman, Multiple View Geometry in Computer Vision, Cambridge University Press, 2000. Derived and summarized relevant theories and algorithms comprehensively and deeply. For different situations where the camera has been calibrated or not, the method of recovering the unknown 3D scene structure ...

AI Technical Summary

Problems solved by technology

However, there are defects in this method: 1) Assuming that the matching error rate is very high, the RANSAC algorithm eliminates few false matches, and the calculated fundamental matrix is ​​unreliable; 2) The best matching pair is used as the initial camera pair for SfM, regardless of the degradation It is very easy to lead to ill-posed estimation; 3) The restored geometric information is sparse three-dimensional feature points, which is far from meeting the visual needs

This method fills the gap between the sparse method and the dense method, and increases the width of a well-matched baseline, but it is also not suitable for complex situations. Only when the internal parameters of the camera change small, the two views of the larger baseline can be obtained. Better quasi-dense matching results; when the camera internal parameter transformation is large, it may get worse results than sparse matching; in addition, this method has a large time overhead

[0007] The method in Document 1 is aimed at the sparse reconstruction of sequential images; Document 2 studies the sparse reconstruction method based on unordered images, and the obtained point model does not have visual expressiveness, and the reliability of the recovered camera pose is poor; Document 3 compares the method of Document 2 Improvements have been made to obtain a better sparse scene model, but it has a huge time overhead; the method in Document 4 can reconstruct a high-density scene model, but it is suitable for sequence images

Images