Use of Lidar Data to Constrain the Matching of Conjugate Features in Large-scale Imagery over Urban Areas

Reliable and accurate 3-D reconstruction of man-made objects, especially buildings, is essential for many applications that involve the use of digital 3-D city models. This paper aims at developing a semi-automated technique for building rooftop reconstruction employing a combination of large-scale aerial imagery and airborne laser scanning data acquired by Light Detection and Ranging (LiDAR) technology. First, the LiDAR data are segmented to derive above-ground planar patches, which are tentatively used as building rooftops. Second, the planar patches are used to obtain rough estimates of the rooftop boundaries. Line extraction and stereo matching techniques are then utilized to obtain actual boundaries of building rooftops from imagery. In this study, Canny Edge Detection together with edge linking and splitting techniques are used to define a group of straight line segments in the imagery. To increase the reliability and accuracy of the matching of conjugate linear features, several constraints are used for candidate line selection in 2D overlapping imagery and 3D LiDAR planar patches respectively, so as to remove some non-probable candidate line pairs. Additionally, a cost function is defined to obtain the correct line pair for each roof boundary. Manual editing is introduced into the final step of the process to deal with cases where the derived boundaries of building rooftops are not complete or not correct. Finally, the performance of these developed procedures is illustrated through experiments using real data.