3D Modeling of Indoor Environments by a Mobile Robot with a Laser Scanner and Panoramic Camera


Peter Biber(1), Henrik Andreasson(2), Tom Duckett(2), Andreas Schilling(1)

mailto: biber(at)gris.uni-tuebingen.de

(1) University of Tübingen, WSI/GRIS

(2) Örebro University, Dept. of  Technology

 


 

Have also a look at our new project for recording 3D models of environments in higher quality: The Wägele

 


This project is concerned with building a 3D model of indoor office environments using a mobile robot. This is a view of a VRML model, that is available for download.

ActivMedia Peoplebot 

It is equipped with a SICK LMS 200 laser scanner and a panoramic camera consisting of an ordinary CCD camera with an omni-directional lens attachment (NetVision360 from Remote Reality).


The route to the 3D model has four stations: Calibrating the sensors, recording the data, generating a 2D map and finally generating textures. The dataflow of the recorded and processed data is shown on the left.

While recording data, the robot is controlled remotely. Whenever its pose changed by a distance of at least 50 cm or an angle of 15 degree, a new scan and a new panoramic image is recorded.


A 2D map is generated from laser scans and odometry by scan matching techniques (Click on the map for a larger image). This map is used to estimate the poses of the robot and to extracts walls, which are later textured.


Textures for the floor are generated by warping the panorama image to the ground plane (according to the calibration) and cropping according to the laser range scan. Floor textures from single images are fused into larger textures by multiresolution blending (see next part).


The resulting model can be downloaded here (zipped vrml). You can view the model for example by using CosmoPlayer (www.karmanaut.com/cosmo/player/)

For a preview, here are some snapshots (click on the images!)


Work in progress: the model will be enriched with additional 3D geometry using the output of a graph cut stereo algorithm. For details see:

www.gris.uni-tuebingen.de/~sfleck/omni3d

and here are two results from the stereo algorithm: