Three-dimensional object model classification method based on multi-view image

Abstract

A three-dimensional object model classification method based on a multi-view image comprises the following steps: 1) rendering three-dimensional object point cloud format data based on a von-von illumination reflection model to obtain a three-dimensional object multi-view image; 2) randomly selecting a plurality of three-dimensional object instances in each category, copying the image obtained inthe step 1 corresponding to the three-dimensional object instances, and expanding the data volume of the corresponding category to obtain a training data set with balanced data distribution; 3) randomly selecting continuous visual angle images from the training data set, and inputting the continuous visual angle images into a pre-trained three-dimensional convolutional neural network to train a target data set; and 4) the size of the three-dimensional convolutional neural network convolution kernel is adjusted, so that the model has a better classification effect. According to the method, thethree-dimensional convolutional neural network algorithm is adopted to carry out feature learning on the multi-view image of the object, the generalized feature of each type of object is effectively learned under the condition that only the continuous view angle of the object exists, the obtained feature has better discrimination, and better classification precision is achieved.

Description

technical field [0001] The invention relates to the fields of deep learning and computer vision, in particular to a method for classifying three-dimensional object models based on multi-view images. Background technique [0002] With the rapid growth of 3D data, research on 3D object data has become more and more important in the field of computer vision. Inspired by the wide application of deep learning models represented by convolutional neural networks (Convolutional Neural Networks, CNNs) in two-dimensional images, three-dimensional convolutional neural networks (3D Convolutional Neural Networks, 3D CNNs) in three-dimensional object classification, recognition, detection , segmentation and other fields have been applied to varying degrees, and achieved good results. [0003] For the classification of 3D objects, there are currently two types of mainstream methods: one is to voxelize the data in the point cloud format and then apply a three-dimensional convolutional neur...

AI Technical Summary

Problems solved by technology

[0005] In order to overcome the shortcomings of existing technical solutions that cannot classify objects based on multi-view images, the present invention proposes a method for effectively classifying objects based on multi-view images, using a three-dimensional convolutional neural network algorithm for feature learning, and the obtained features are Better discriminative, with better classification accuracy

Images