Extraction of building roof contours from the integration of high-resolution aerial imagery and laser data using Markov random fields

Abstract

ABSTRACTThis study proposes a method for the automatic extraction of building roof contours from the integration of airborne laser scanning (ALS) and photogrammetric data using the Markov random field probabilistic approach. Initially, a normalised digital surface model (nDSM) obtained from the ALS point cloud is segmented to obtain the polygons that represent high objects in the scene. These polygons are projected onto the image to delimit regions (sub-images) that will be segmented in the image, which allows the extraction of polygons in the image that represent the corresponding high objects. The polygons that represent roof contours are identified by the optimisation of an energy function that models the geometric and contextual properties of building roof contours via the genetic algorithm. This energy function combines the polygons extracted from the nDSM and from the image. The proposed method was evaluated with real data, including high-resolution aerial images and ALS data. The experimental results showed that the proposed method works properly, exhibits few failures and has average completeness and correctness rates above 90%.