Conclusion

This paper has presented a new approach to sensor and knowledge based reconstruction of 3D indoor environments with autonomous mobile robots equipped with a 3D laser scanner. The proposed method consists of three steps and is applied after the 3D data is acquired:

• The first step is a fast feature extraction, i.e., plane detection. The presented algorithm is a combination of the ICP algorithm with the RANSAC approach.
• Second, the computed planes are labeled with a predefined semantic net. The semantic net contains and implements general knowledge of indoor scenes. The semantic net is externalized and implemented as a Prolog program. 3D analysis of the extracted features compiles additional clauses and Prolog's backtracking and unification algorithm derives scene specific knowledge.
• Third the model is refined with the constraints arising from the semantic labeling. Numerical algorithms, i.e., Powell's method and the downhill simplex method are used for the 3D model improvement.

The proposed method will be included in the robot control architecture for the automatic gaging of indoor environments.

Future work will concentrate on the integration of two color cameras and enhancing the semantic interpretation by fusing color images with range data. The aperture angle of the camera will be enlarged using a pan and tilt unit to acquire color information for all measured range points. Furthermore the semantic net will be extended to more detailed features, i.e., non-planar features.