Conclusions

This paper has presented a novel method for the learning, fast detection and localization of instances of 3D object classes. The 3D range and reflectance laser scanner data are transformed into images by off-screen rendering. For fast object detection, a cascade of classifiers is built, i.e., a linear decision tree [25]. The classifiers are composed of classification and regression trees (CARTs) and model the objects with their view dependencies. Each CART makes its decisions based on feature classifiers. The features are edge, line, center surround, or rotated features. After object detection the object is localized using a point matching strategy. The pose is determined with six degrees of freedom, i.e., with respect to the $x$, $y$, and $z$ positions and the roll, yaw and pitch angles. A final computation returns a quality measure for the object localization.

The presented combination of algorithms, i.e., the system architecture enables high accurate, fast and reliable 3D object localization for autonomous mobile robots.