Building and Road Extraction from LiDAR Data
This chapter starts with a generic context-based classification technique that assigns semantic class labels to each Light Detection and Ranging (LiDAR) point. To do so, intermediate steps, such as calculation of feature vectors, the selection of classifiers, and consideration of the relations between neighboring LiDAR points or segments, and formation of graph structures are discussed in detail. Having classified the LiDAR points, a probabilistic context-based technique for building detection on the basis of the labeled point cloud is described. The subsequent stage consists of the reconstruction of polyhedral models from the point cloud data, including a method for the consistent estimation of all model parameters and regularization. Treated as a classification problem, road detection is carried out by a rule-based binary classifier. Extracted road segments are vectorized to determine road centerlines, which are then combined to form the road network. LiDAR classification and building extraction are demonstrated using the International Society of Photogrammetry and Remote Sensing (ISPRS) Test Project on Urban Classification and 3D Building Reconstruction, whereas building reconstruction and road extraction are evaluated for a dataset from Australia.