Regional Invariant Descriptors in Mobile Robotics Part II: Enhanced Detection and Matching

In Part I of this research work [1] we analysed the stability tracking of three well-known invariant descriptors (i.e., SIFT, MSER and QUICK-SHIFT). The MSER algorithm resulted to be less vulnerable than the other algorithms to fail detecting feature points. However, its instability due to sensor uncertainty and light variations was similar to other algorithms. In this paper, we now propose a novel method that improves the invariants stability detection to reach a state called binary stable region as an input image to apply MSER. The method is composed by a set of filtering algorithms that automatically select natural environmental features, which are subsequently transformed into invariant regional descriptors, to be tracked as steady environmental landmarks. Indoors/outdoors experimental results with a mobile robotic platform are presented to show the robustness and reliability of the proposed algorithm.