Tracking Spatially Distributed Features in KLT Algorithms for RGB-D Visual Odometry

Mobile robots can operate without any human intervention in the most varied scenarios provided they are endowed with autonomous capabilities. One of these capabilities is to localize themselves within the environment of operation, which can be performed by Visual Odometry, a process that estimates relative camera motion based on information extracted from images. In this work, we propose feature tracking algorithms for RGB-D (e.g. Kinect) Visual Odometry. We design our solution based on the hypothesis that more accurate motion estimates are achieved when visual features are spread uniformly on images. Experimental results demonstrate that this heuristics is indeed valid and presents the performance of our algorithms on standard datasets.