3D digitalization of environments without occlusions requires multiple 3D scans. Autonomous mobile robots equipped with a 3D laser range finder are well suited for gaging the 3D data. Due to odometry errors, the self localization of the robot is an unprecise measurement and therefore cannot be used for registration of the 3D scans in a common coordinate system. The geometric structure of overlapping 3D scans has to be considered. Our approach uses a newly developed, fast version of the well known Iterative Closest Point (ICP) algorithm, a method for aligning three dimensional models purely based on the geometry.
To build complete volumetric models, multiple 3D scans have to be registered. Most published registration methods concentrate on pairwise alignment of two 3D scans. Pulli concludes that extending these pairwise methods for a multiview case has proven not to be straightforward, since simply chaining pairwise registration over all scans seldom works [13]. The goal of the work presented here is to develop a method that does work. To acquire the multiple 3D scans a robot equipped with the AIS 3D laser range finder explores the world and creates reliably a precise and consistent 3D volumetric representation, in real-time.
Instead of using 3D scanners, which measure intrinsic
consistently the environment, some groups have attempted to build
3D volumetric representations of environments with 2D laser range
finders. Thrun et al. [10,21], Früh et
al. [8] and Zhao et al. [23] use two
2D laser range finder for acquiring 3D data. One laser scanner is
mounted horizontally, the other vertically. The latter one grabs
a vertical scan line which is transformed into 3D points using
the current robot pose. Since the vertical scanner is not able to
scan sides of objects, Zhao et al. use two additional vertical
mounted 2D scanner shifted by 
to reduce occlusion
[23]. The horizontal scanner is used to compute the
robot pose. The precision of 3D data points depends on that pose
and on the precision of the scanner.
A few other groups use 3D laser scanners [16,1]. A 3D laser scanner generates consistent 3D data points within a single 3D scan. The RESOLV project aimed to model interiors for virtual reality and tele presence [16]. They used a RIEGL laser range finder on robots and the ICP algorithm for scan matching [4,6,22]. The AVENUE project develops a robot for modeling urban environments [1], using a CYRAX laser scanner and a feature based scan matching approach for registration of the 3D scans in a common coordinate system [18]. The research group of M. Hebert has reconstructed environments using the Zoller+Fröhlich laser scanner and aims to build 3D models without initial position estimates, i.e., without odometry information [11].
The paper is organized as follows. Sections 2 and 3 describe the used 3D laser range finder and the mobile robots. In section 4 we present the scan matching, followed by the application of matching sewer pipes in section 5. Section 6 concludes the paper.