3D Model Recognition Based On Machine Learning

3D modeling is an important research branch of computer graphics.With the rapid development of 3D data acquisition equipment and the improvement of computer performance,3D model is more and more widely used in scientific research,life and work.More and more researchers are devoting their attention to the field of 3D model recognition.In this paper,some methods of machine learning are used to study the problem of 3D model recognition.The main work of this paper is as follows:Firstly,considering that the shape of the 3D model is mainly determined by the normal direction of the 3D object surface,an area-weighted normal direction histogram feature for3 D model recognition is designed.The algorithm is validated on the Model Net10 database provided by Princeton University.The experimental results show that the proposed algorithm has higher recognition rate than the existing algorithms such as SPH,LFD and 3D Shape Net.Secondly,an algorithm for 3D model recognition based on multi-view projection and SIFT features is proposed.By observing the 3D model from different angles and projecting it onto different planes,multiple views of the 3D model are obtained.Then,SIFT features,which are widely used in the field of 2D image recognition,are extracted from each view and one class label is obtained for each view.Finally,we use voting mechanism to fuse the recognition results of each view and the final recognition result is obtained.The algorithm is verified by numerical experiments on the Model Net40 database provided by Princeton University,and different methods are compared.Finally,a 3D model recognition algorithm based on SIFT feature and sparse coding is proposed.Sparse coding of SIFT features is carried out by using feature symbol search algorithm and Lagrange dual method.Spatial pyramid method is used to pool the sparsely coded features to obtain the feature vectors of the 3D model,and then support vector machine is used for classification and recognition.Experiments on Model Net40 database show that the accuracy of this algorithm is higher than most existing algorithms.