Grading-based positive and inverse reversing three-dimensional dense point set rapid registering method

Abstract

The invention provides a grading-based positive and inverse reversing three-dimensional dense point set rapid registering method which comprises the following steps of: giving two point sets, wherein one point set is a reference point set, and the other point set is a target point set; solving two transformation functions, wherein the positive transformation function h is a transformation function from the reference point set to the target point set, and the inverse transformation function g is a transformation function from the target point set to the reference point set; leading h to be approximately equal to g<-1> and leading g to be approximately equal to h<-1>, wherein h<-1> and g<-1> respectively represent inverse functions of h and g; and finally obtaining a precise corresponding relation of the two point sets by the two transformation functions. Compared with the prior art, the method has the advantages of registration based on a grading way of global registration and local registration, defines an inverse consistency error and a characteristic point for local registration by utilizing a difference of the corresponding relations in the positive direction and the inverse direction, prevents complex and time-consuming function inversion operation in the traditional positive and inverse reversing registration algorithm and has simple realization, high efficiency and high registering precision.

Description

technical field [0001] The invention belongs to the field of computer applications and relates to a three-dimensional dense point set-to-point set registration method for computer vision and pattern recognition. Background technique [0002] Registration is a basic problem, which originates from many practical problems in many fields, such as the fusion of information obtained by different sensors; the difference detection of images or models obtained at different times and conditions; Three-dimensional information acquisition; pattern or target recognition in images, etc. Although the objects involved in the registration can have various forms, such as the geometric model of the 3D object, 2D or 3D images, its essence is the registration problem of two 2D or 3D point sets, that is, to find the distance between the two point sets. corresponding relationship. [0003] The iterated closest point algorithm (Iterated Closest Point, ICP) is the most commonly used point set regi...

AI Technical Summary

Problems solved by technology

However, almost all existing methods involve solving the inverse function. Due to the discrete nature of the processing objects, the solving process takes a long time, and cannot be applied in many fields and industries with real-time requirements.

Although there have been some studies on how to improve the efficiency of forward and reverse reciprocal registration, the current results cannot meet the requirements of practical applications.

Images