SEGMENTS: a layered, dual-Kalman filter algorithm for indoor feature extraction

A layered algorithm for extracting features in an indoor environment from planar range data is presented. At the lower (signal processing) level, the SEGMENTS algorithm exploits the speed and accuracy of two extended Kalman filters working in tandem and processes sequentially the distance measurements within each scan. The first Kalman filter is responsible for initiating straight line segments and detecting clutter while the second estimates the parameters of each line segment (such as distance and orientation) and determines when this is interrupted. The dual filter combination is capable of detecting edges and straight line segments within the scene in front of the robot. At the higher (post-processing) level, the identified segments are combined to form more complex features such as extended walls, corners (concave and convex), doors and corridors. During the composition cycle, SEGMENTS makes use of (i) the parametric representation of the straight line segments and (ii) the prespecific topological models of the features this algorithm seeks. A list of the identified features along with their location with respect to the laser sensor is finally available to the user. The cluttered regions in the scene are also marked on a polar representation of the environment. The presented algorithm has been tested on a Pioneer 2 DX mobile robot equipped with a SICK LMS 200 proximity laser scanner performing map-based localization. Low computational requirements, accuracy and robustness to uncertainty and noise characterize the performance of this new method for feature extraction