3-D Point Cloud Recognition Based On Spherical Harmonic Function

Object recognition is a core technology to realize scene understanding in the field of computer vision.Traditional object recognition obtains object category information by analyzing two-dimensional images.Because image acquisition varies with illumination,attitude and other factors,the reliability of object recognition technology based on two-dimensional images is not guaranteed.The method of describing the shape of an object with three-dimensional point cloud data can accurately and comprehensively describe the geometric information of the object,and has the invariance of illumination and attitude.Especially with the gradual development and maturity of lidar,depth sensor,coordinate measurement system and image-based three-dimensional reconstruction technology,the acquisition of three-dimensional point cloud data is becoming more and more convenient.Therefore,the research of object recognition technology based on three-dimensional point cloud has also received extensive attention and become a hot research direction in the field of computer vision.This technology will also greatly promote the commercial application of automatic driving,intelligent robots,intelligent medical,video surveillance,virtual reality and augmented reality.In this paper,an efficient recognition algorithm of point cloud based on local spherical harmonic function is proposed,aiming at the characteristics of sparse point cloud data and easy to generate holes due to low resolution of equipment,complex environment and three-dimensional reconstruction method based on image features.Firstly,a maximum eigenvector based on the covariance matrix constructed by three-dimensional point cloud is proposed to realize fast alignment between the object to be measured and the standard model;secondly,the feature points of three-dimensional point cloud are defined based on the surface curvature extremum defined by point cloud,and the corresponding spatial regions are defined and divided according to the matched feature points;secondly,the corresponding sub-regions are obtained according to the spherical harmonic function expansion.The coefficients corresponding to the basis functions constitute the coefficients describing method based on spherical harmonic function expansion for the shape of the corresponding sub-regions,and then construct the similarity comparison method between the two corresponding sub-regions.Finally,the similarity of several sub-regions is counted to achieve the similarity comparison of the whole object.95.1% recognition accuracy is obtained on the basis of Princeton's shape reference library,and 92.9% recognition accuracy is obtained on the expansion of UWA database.Experiments show that the algorithm has obvious advantages over the existing algorithms in computational efficiency.