Efficient model detection in point cloud data based on bag of words classification

To detect point cloud data of models is an important problem in computer vision. In this paper, an approach based on the bag of words model and the naive Bayes classifier is proposed for detecting point clouds models. The approach first uses a moving-least-squares (MLS) technique to calculate the geometric information and select the salient points to extract necessary features. The point feature extraction is achieved by using spin image signatures. After that, a bag of words model is introduced to form the global feature for each model. Finally, a naive Bayes classifier is trained to detect models across the database. To conclude, the experimental results show that our approach outperforms the original spin image signature method, and has good performance in partial model detection. Moreover, the timing shows that the proposed approach is suitable to detect models efficiently.