Unsupervised plane data clustering based on modified dirichlet process mixture model method

Terrestrial laser data acquisition for 3D Urban Modelling is becoming common. Processing the relatively large amount of data in the acquired point clouds is expensive in computational time and memory. In this paper, we provide a solution: to process the raw point clouds into planes by data classification with multi-scale Conditional Random Field, followed by data division and plane patches fitting for large scale data. To group the plane patches into locally delimited planes, we proposed using the Dirichlet Process Mixture Model (DPMM). We modified the Gaussian mixture function in the DPMM to optimise plane fitting, and we demonstrated the efficacy of the algorithms on two sets of real world data. The result showed the proposed method is more robust compared to the previous work for both plane data and plane patches unsupervised clustering.