3D registration method based on local deformation of 2D images

Abstract

The invention relates to a three-dimensional registration method based on local deformation of a two-dimensional image. The method comprises three main steps including preliminary registration, localdeformation and precise registration. The method comprises firstly, adjusting a three-dimensional model scene to be consistent with a two-dimensional image scene, calculating a projection matrix by selecting an image coordinate point corresponding to a three-dimensional coordinate, and projecting the two-dimensional image onto the three-dimensional model to complete the preliminary registration; then selecting the coordinates, erroneously registered with the three-dimensional model, in the two-dimensional image, moving the erroneous coordinates to correct positions by moving least squares method local deformation; and finally calculating a coordinate mapping table of all regions of the two-dimensional image or the region targeted by the deformation, readjusting the two-dimensional image byusing the coordinate mapping table, and projecting the new two-dimensional image onto the three-dimensional model to complete the final registration. The method is beneficial to reducing the error caused by the preliminary registration, especially the local error, thereby obtaining a better three-dimensional registration effect.

Description

Technical field [0001] The invention relates to a three-dimensional registration method based on local deformation of a two-dimensional image. Background technique [0002] Three-dimensional registration is the matching and fusion of basic information in a two-dimensional image (two-dimensional coordinates and pixels of points, lines, surfaces, etc.) with three-dimensional spatial information (three-dimensional coordinates and textures of points, lines, and surfaces, etc.), which requires quantitative registration Analyze the size and position relationship of space objects. The existing three-dimensional image registration methods can be divided into three main categories: [0003] 1) Learning method based on feature statistics. This method is based on a large target database. The target in the database is compared with the actual target, the probability function is established, and the target depth is registered through the probability. At present, the probability models that h...

AI Technical Summary

Problems solved by technology

The registration method based on the shape recovery technology requires a lot of light and grayscale, and needs to preprocess the image's luminosity and grayscale, and the image resolution and foreground and background grayscale differences will affect the registration effect; and this item If there are no other constraints, the technology cannot uniquely match curved objects such as convex and concave surfaces.

The 3D registration technology of a single image based on geometric projection information has achieved a relatively ideal effect on the target image with geometric structures, but the accumulation of measurement errors is large, and the deformation of the 2D image affects the local registration effect. If it can be better eliminated These defects, this method can be regarded as a more adaptable registration method with simple operation, fast speed, less investment and high efficiency

Images