Abstract: Studies of building types play an important role in the earthquake rescue and damage assessment, but now research of building classification is relatively few based on dense point clouds of SfM (Structure from Motion) method. Firstly, this paper use SfM method to generate dense points, and classify the buildings height and the roof types with the mean and standard deviation of the height of dense point clouds, and then use the factor of the plane distance of the averages of high and low points to classify the roof types. The study area of this paper is Maqu country, and we processed the UAV images using SfM method and used the factors to classify the height and roof types. For the dense point clouds data from SfM method, the experimental results show that when setting the threshold of height and standard deviations of 6 m and 0.25 m respectively can accurately distinguish the layer number of building and the roof. Using the distance factor to distinguish the shed and dual slope construction, when the distance threshold is 1.5 m, we will get ideal result. Classification of typical buildings with dense point clouds can provide a certain reference function for earthquake disaster risk analysis and future earthquake disaster loss assessment.

Key words: Building roof geometry feature, SfM, 3D point clouds, Classifying factors