A Method for Extracting Local Spherical Harmony Feature of 3D Model

Abstract

The invention discloses a three-dimensional model local spherical harmonic feature extraction method and belongs to the field of computer vision. The three-dimensional model local spherical harmonic feature extraction method is characterized in that the local spherical harmonic feature extraction method is provided in allusion to the defect that in the existing three-dimensional model global spherical harmonic feature extraction method, the selection deviation of the centre of sphere has a strong impact on the subsequent spherical surface sampling and feature extraction. The method comprises the steps of performing dimension normalization processing on a three-dimensional model; performing resampling on three-dimensional data by using an interpolation method so as to generate a depth image, and performing mesh generation on the depth image so as to determine positions of local centers of sphere; performing local spherical surface sampling on each local center of sphere by using a homocentric spherical shell method; calculating a spherical harmonic feature vector corresponding to each local center of sphere, and using a weighted array method to obtain local spherical harmonic feature vectors of the whole three-dimensional model. By using the method, the instability caused by the position deviation of the centre of sphere can be relieved to some extent, and an excellent basis is provided for subsequent recognition.

Description

technical field [0001] The invention belongs to the field of computer vision, and relates to spherical harmonic transformation, three-dimensional model feature extraction and other technologies. Background technique [0002] Summarizing the existing 3D model recognition methods, there are two main methods of feature extraction: one is the feature extraction method based on the depth map, and the other is the method based on the 3D data itself. The feature extraction method based on the depth map first corrects the pose of the 3D model, then generates the depth map, and finally uses the 2D feature extraction method to perform feature extraction on the depth map. Although the depth map can represent the shape characteristics of 3D data at a certain level, it is a depth projection in a certain aspect after all, so the selection of the projection direction is a very important process. In general, it is necessary to analyze the 3D model before projection. Perform attitude rotati...

AI Technical Summary

Problems solved by technology

It can be seen that although it is called the 3DLBP algorithm, the calculation is not really carried out in 3D data, but is applied to the depth image of the 3D image. Although the depth image can represent the structural information of the 3D image, the depth image is 3D. The projection result in a certain direction of the image will inevitably cause the loss of some structural information, so it should actually be called 2.5D LBP algorithm

Images