Segmentation and parameterization of 2D lines based on Mean Shift Clustering

This paper presents a robust algorithm for segmentation and characterization of lines detected by a laser sensor. We propose a strategy of Mean Shift Clustering which using the points of the laser scan performs a classification stage based on an ellipsoidal orientable window previous to the line segment parameterization. Each data set is processed by a RANSAC (Random Sample and Consensus) algorithm modified to detect spurious. This method reduces the amount of spurious and updates associated probability densities. The parameters of the detected segments are estimated by TLS (Total Least Squares) regression. The algorithm has been evaluated in indoor environments using as mobile platform the robot Pioneer 3DX equipped with a SICK laser.