Unsupervised texture segmentation for 2D probabilistic occupancy maps

This paper presents a novel method for the segmentation of probabilistic two-dimensional occupancy maps, based on the analysis of their texture characteristics. The texture is represented by means of a double distribution of "local binary pattern" and "contrast". The logarithmic likelihood ratio, G-statistic, is used to measure the degree of similarity between different regions; this pseudo metric measure compares LBP/C distributions linked to different segments. The innovative algorithm is used to segment the probabilistic images in regions that characterize the space according to the certainty of its occupancy level. For a better interaction between an autonomous system and its environment, the segmentation scheme is also able to differentiate between objects present in the scene by analyzing the proximity between occupied segments. Along with experimental results, a comparison with other algorithms is provided in order to demonstrate the efficiency of the proposed approach