Building Extraction from Point Cloud Using Marked Point Process

Objective In this paper,a marked point process based method is used to extract buildings from air-borne LIDAR data.At first,a Gibbs energy model is build according to the geometric feature of theobject in the point cloud data.This model contains both a data coherence term which fits the objects tothe data and a prior term which incorporates the prior knowledge of the object geometric properties.Then the previously defined model is optimized by the RJMCMC(Reverse Jump Markov Chain MonteCarlo)algorithm and simulated annealing algorithm.Finally,fine processing removes the terrestrialpoints,noise points and tree crown points of the extracted objects which are mistakenly extracted asbuildings,while combining adjacent objects.The method was tested with three different aerial LiDARdata sets from ISPRS.The results show that our method is capable of efficient and robust building ex-traction.