A method for automatic recognition of various geometric primitives in 3D point cloud

Abstract

The invention discloses a method for automatically recognizing various types of geometric elements in a three-dimensional point cloud, and the method comprises the steps: carrying out the preprocessing of an inputted three-dimensional point cloud, i.e., voxel filtering, building a neighbor structure based on a Kd tree, and estimating a normal vector of points; carrying out the determining of a neighbor relation of the point cloud, and carrying out the sampling of the point cloud; calculating a covariance matrix of sample point neighbor, analyzing the size relation among three characteristic values, and generating a corresponding initial geometric element model according to a coplanar rule; respectively building a corresponding energy equation according to the initial geometric element models, and carrying out the planar, spherical and cylindrical surface energy calculation according to an energy optimization frame; carrying out the loop iteration of the above steps, minimizing the energy of various types of geometric elements, solving geometric element parameters under optimal significance through employing an optimization algorithm, thereby achieving the refining of the parameters of geometric element models; and finally outputting the parameters and inner points of a plurality of types of geometric elements. According to the technical scheme of the invention, the method is wide in application range, is accurate in parameter estimation, is strong in anti-interference capability, and greatly improves the recognition and analysis capability of the three-dimensional point cloud.

Description

technical field [0001] The invention relates to the technical field of three-dimensional perception of computer vision and robot navigation, in particular to a method for automatic detection and recognition of objects in a three-dimensional point cloud. Background technique [0002] In recent years, computer vision research has been booming. Researchers have continuously created groundbreaking algorithm theories and designed new product technologies, thus endowing machines with visual capabilities closer to human beings and bringing great benefits to people's lives and work. Earth-shaking changes. In particular, the popularity of high-performance camera and photography equipment, the rapid increase in computer computing speed, and the breakthrough of algorithm theory with learning ability have made computer vision play an important role in various fields such as robotics, security monitoring, industrial production, game entertainment, and medical imaging. . RGB-D, a high-p...

AI Technical Summary

Problems solved by technology

But then a challenging problem arises: how to use the computer to automatically analyze and perceive the huge 3D data collected

This type of method handles a single model more efficiently, but some models will be missed when fitting multiple models; it is very dependent on the distance threshold, and usually needs to be manually adjusted continuously; and it cannot analyze the relationship between each model from a global perspective. Attribution of interior points

The region growing algorithm needs to manually select some internal points in advance and then expand the growth. It cannot be analyzed automatically by the computer, and it is very sensitive to external points and noise, and its robustness is poor.

The main problems of these methods are: first, they cannot identify multiple geometric primitives at the same time; second, they have poor anti-interference ability to external points and noise; third, they rely on angles and thresholds to judge the internal points of geometric primitives, and the degree of automation is low.

Images