Regional Invariant Descriptors in Mobile Robotics Part I: Stability Tracking Analysis

Visual landmark tracking represents a major problem due to occlusion, illumination variations and affine transformations between subsequent images. A desired goal for several mobile robot applications is to increase a steady state in the invariants in order to minimize errors in the tracking. This manuscript consists of two parts. The first part presents a quantitative analysis of stability tracking of invariant descriptors. The analysis starts by comparing two feature extraction algorithms: the Harris Corner Detection and the Fast Corner Detection. Then, a comparative study of three known invariant algorithms (SIFT, MSER, Quick SHIFT) is carried out to track multiple regional descriptors correlated among consecutive video frames. These algorithms were compared on faults tolerance and computational complexity. From the results obtained in this analysis, in Part II of this paper, we propose a novel method that improves the invariants stability detection.