Regular scene three-dimensional information extracting method based on structure prior

Abstract

The invention discloses a regular scene three-dimensional information extracting method based on structure prior. The method comprises the following steps of utilizing PMVS algorithm to produce three-dimensional space directional point cloud; performing classifying process on the direction distribution of the three-dimensional direction point by an unsupervised clustering algorithm, obtaining a space significant plane normal direction; calculating the position offset of the three-dimensional space point cloud on the plane normal direction according to the coordinate of the three-dimensional space point cloud and the plane normal direction thereof, and clustering the position offset of the three-dimensional point cloud to obtain the offset volume of the space significant plane; respectively pre-segmenting each two-dimensional image, smoothing the edges of the segment region, and obtaining the regional segment information; establishing a Markov random field model by using the three-dimensional space significant plane as the annotation and using the regional segment information of the two-dimensional images as the node, and solving the model based on a global optimization method, so as to obtain the segmented coplanar depth information in each viewpoint image. The method provided by the invention can be used for realizing higher accuracy and lower calculation complexity.

Description

technical field [0001] The invention relates to a method for extracting three-dimensional information, in particular to a method for extracting three-dimensional information of a regular scene based on structure prior, which is used for three-dimensional structure reconstruction of a regular scene based on multi-viewpoint images, and belongs to the technical field of image processing. Background technique [0002] Image-based 3D scene reconstruction refers to the use of images captured by cameras and other equipment as input, and the comprehensive use of various technologies to study the 3D coordinates of 3D structural primitives such as points, lines, and surfaces in 3D space and the corresponding 2D points in 2D images. , the relationship between the two-dimensional coordinates of lines and surfaces, and the technology to determine the three-dimensional information of objects. After years of innovation and development, it has become one of the important research contents o...

AI Technical Summary

Problems solved by technology

Therefore, when reconstructing a 3D scene based on a multi-view stereo algorithm, the feature matching relationship between each viewpoint is extremely limited, so that it cannot provide sufficient constraint information for the 3D geometric space.

[0005]2) There are relatively serious occlusions between the images of each viewpoint

[0009] Traditional 3D information acquisition technology is not ideal for reconstruction of regular scenes

To sum up, the shortcomings of the existing 3D scene reconstruction information extraction based on structure prior mainly include: traditional methods have relatively strict model assumptions for regular scene reconstruction, such as the Manhattan-world stereo model, which only assumes that the three-dimensional plane direction is three Approximate mutually perpendicular axis directions, poor universality

Traditional methods do not make full use of a large number of highly regular 3D structures (such as collinearity and coplanarity) in regular scenes, and only use discretized depth values ​​as labels, and use pixels as nodes to model energy cost functions, that is, only use relatively Small-scale primitives are structurally constrained, so that the obtained 3D information is not accurate enough

Images