Quasi-Thematic Feature Detection and Tracking For Future Rover Long-Distance Autonomous Navigation

This paper investigates state-of-the-art approaches for object detection and tracking employing models that can efficiently detect objects (specifically ‘rock’ on planet surfaces) in the visual scene in terms of semantic descriptions. Two models (i.e., “visual saliency” and “blob (shape-based) detection”) are presented here specifically focused towards future
planetary exploration rovers. We believe that these two object detection techniques will abate some of the algorithmic limitations of existing methods with no training requirements, lower computational complexity and greater robustness towards visual tracking applications over long-distance planetary terrains. Comprehensive (quantitative) experimental
analysis of the proposed techniques performed using three challenging benchmark datasets (i.e., from PANGU, RAL Space SEEKER and SSC lab-based test-bed) will be presented in this paper.