A Robust Registration Method for the Reconstruction of 3D Urban Objects using Ground-based Laser Range Scanner

• Published in: Journal of the Japan society of photogrammetry and remote sensing Vol. 39; no. 2; pp. 52 - 63
• Main Authors: ZHAO, Huijing, SHIBASAKI, Ryosuke
• Format: Journal Article
• Language: English
• Published: Japan Society of Photogrammetry and Remote Sensing 2000
• ISSN: 0285-5844; 1883-9061; 1883-9061
• DOI: 10.4287/jsprs.39.2_52

There is potentially strong demand for detailed 3D urban spatial data especially for advanced automobile/pedestrian guidance system and radio disturbance analysis for telecommunication and so forth. Ground-based laser range scanner is one of the promising devices to efficiently acquire detailed 3D spatial data such as facade of buildings and detailed urban objects along streets. From a single laser range image, however, entire surface of an urban object can not be covered due to occlusions. Multiple range images acquired from different locations and with different angles have to be registered with each other to minimize occlusions and to generate more complete 3D models. In this paper, a robust method for registering laser range images of an urban object like buildings is proposed under the assumption that the laser scanner is located so that the rotating axis is vertical to the ground. Registration is achieved in two steps; a pairwise registration of successive range images and simultaneous registration of multiple range images to adjust errors accumulated in the pairwise registration process. Pair-wise registration method to determine 4 transformation parameters, a horizontal rotation angle and three translation parameters is based on “Z-image”, which is generated by projecting 3D points computed from a range image onto the horizontal (x-y) plane. In the simultaneous registration of multiple images, accumulation of estimation error in pairwise registration is adjusted with a least square minimization method. An outdoor experiment is conducted to reconstruct a 3D model of the buildings in the campus of the University of Tokyo using twenty-nine range images. With the registration method, around twenty-three pairs of range images were automatically registered, while six pairs were registered with a manually assigned initial transformation. A 3D model of the buildings was generated in an automated mode. Accuracy of the model was examined using a 1: 500 scale digital map and GPS measured location of viewpoints.