Feature extraction from range images in 3D modeling of urban scenes

Since data acquisition is executed outdoors, the quality of data is not very satisfactory, and special treatments are involved in our extraction methods in order to provide rich and accurate feature information for processing at upper level. In plane extraction, planar cluster-selection & filtering algorithm is proposed to overcome the classic antinomy about size of the window used to test planarity; in line segment extraction, segments grow along the boundaries to overcome the misleading effect of those irregular jump-edge points created by tree leaves. Experiment results verify that the algorithms are quite effective for the range data where much noise from trees, parking cars and pedestrians exists.