Chain feature extraction matching method for three-dimensional registration of mechanical parts

Abstract

The invention discloses a chain feature extraction matching method for three-dimensional registration of mechanical parts. Aiming at the characteristic that the surface of a mechanical part has no texture but has rich geometric information, directed straight line segments connected end to end in a two-dimensional image of the mechanical part form a directed chain, and a feature descriptor is constructed according to the shape of the chain. In the feature matching stage, a dynamic weight is introduced to calculate the similarity distance between feature vectors, so that chain-shaped features with large shape and structure differences are quickly eliminated. In addition, the feature can construct multiple sets of 2D-3D corresponding points between the two-dimensional image and the three-dimensional model of the mechanical part at the same time. The three-dimensional registration efficiency is improved and the robustness of registration is enhanced at the same time.

Description

technical field [0001] The invention belongs to the field of graphics and image processing, in particular to a chain feature extraction and matching method for three-dimensional registration of mechanical parts. Background technique [0002] Fast and robust 3D registration of mechanical parts through monocular images is the basis for realizing an augmented reality assembly application system. The 3D registration based on monocular images only needs an ordinary RGB camera and a CAD model to realize the pose estimation of the target object, and its convenience and quickness have attracted the interest of many researchers. However, mechanical parts have the characteristics of no texture and metallic luster on the surface, which makes some mature image features (such as SIFT, SURF) unsuitable, which increases the difficulty of monocular 3D registration. [0003] At present, the 3D registration of monocular images usually adopts the 3D registration framework based on model key f...

AI Technical Summary

Problems solved by technology

The disadvantage of this image feature is that the method of determining the direction of the line according to the gradient direction of the midpoint of the line segment is sensitive to illumination, and it is easily affected by the noise line segment near the feature line during the feature construction process, which makes the registration robustness drop significantly.

[0005] The above methods have certain deficiencies in registration efficiency, stability, etc., resulting in a large room for improvement in the engineering application process.

Images