A multi-start incremental 3D reconstruction method based on large-scale disordered images

Abstract

The invention discloses a multi-start incremental three-dimensional reconstruction method based on large-scale disordered images, including the steps of image matching and matching relationship diagram construction, reconstruction starting point selection, boundary detection, multi-start three-dimensional reconstruction and sub-model splicing; Class and iterative drift strategy, adaptively select the image pair suitable as the starting point from the unordered image set and start reconstruction at the same time; before the reconstruction starts, determine the best reconstruction sub-process for each image according to the hierarchical shortest path algorithm, determine Reconstruct the boundary; and according to the sub-models obtained by different sub-processes, they are spliced ​​through the common parts they contain to obtain a complete 3D reconstruction model; the 3D reconstruction method provided by the present invention solves the problem of reconstruction when the scene contains multiple non-overlapping parts. The problem of process interruption is to obtain the complete 3D model covered by the image collection; it also avoids the situation that the 3D structure is wrongly transmitted through the wrong image matching; the reconstruction sub-process can be processed in parallel, which improves the image reconstruction efficiency.

Description

technical field [0001] The invention belongs to the technical field of computer vision, and in particular relates to a multi-start incremental three-dimensional reconstruction method based on large-scale disordered images. Background technique [0002] The 3D reconstruction based on computer vision technology can restore the 3D structure of the scene from the 2D image, which conforms to the habit of human beings to recognize the world, and has a wide application prospect. Using existing images on the Internet for 3D reconstruction can greatly simplify the steps of 3D reconstruction data collection, reduce the cost of data collection, and shorten the time of data collection. The parameters are unknown and the overlapping relationship of image scenes is unknown. [0003] The existing large-scale unordered image collection scene 3D reconstruction adopts the method of Structure from Motion (SFM), including four steps of image matching, initial image pair selection, incremental ...

AI Technical Summary

Problems solved by technology

[0005] In view of the above defects or improvement needs of the prior art, the present invention provides a multi-start point incremental 3D reconstruction method based on large-scale disordered images. Problems with Accuracy and Error Accumulation

Images