Similarity Calculation of 3D Building Model based on EMD of Graph Features

In this paper, a feature extraction method based on 3D building topology graph is proposed. By using the model simplification preprocessing method, the main surface and their connection relations of the model are generated, and the feature vector set of the 3D building model is created by calculating its geometric space attributes and topological relations. Then, the earth mover's distance (EMD) algorithm is used to calculate the similarity of 3D building feature vector geometry, so as to complete the clustering of urban 3D models. For different types of models, typification operation can realize rapid model simplification and generate multi-level of detail models, so as to improve the efficiency of visualization and interaction and support the dynamic rendering of VR all-in-one machines and other platforms. Through experimental comparison, this paper proves the effectiveness of the proposed adjacent face feature, and achieves 93% classification accuracy on the standard citygml dataset, which is higher than the traditional MVCNN and other methods.