Robust Unified Optimal Monocular Tracking and Servoing: New Techniques and Applications to Autonomous Systems

Abstract

This document presents a one-year sabbatical research proposal in the field of robot vision. More specifically, the project addresses the challenging problem of optimal monocular tracking and servoing with respect to unknown rigid objects in cluttered dynamic environments. Object unknowns include depth, size, orientation, and texture, whereas cluttered dynamic scenes present moving obstacles and agents, what are likely to occlude the field of view. That is a fundamental problem that arises in many real-world applications of autonomous systems, such as robotic operations, remote sensing, and environmental surveillance. An ideal solution simultaneously requires robustness given the working environment, versatility given the object unknowns, computational efficiency given the real-time constraints, and accuracy in all translations and rotations given the envisaged applications. Existing approaches do not satisfy all these requirements. Also motivated by a lack of Brazilian expertise in such a promising field, this work will be developed with Prof. Dr. Markus Vincze, who is with TU Wien, Austria. His world-class expertise and professional network of nearly 30 years in the domain of robot vision are of paramount importance to help solving that challenging problem. We expect to benefit from the mutual experiences, and propose a complete solution and validate it with real-world thematic experiments. This research proposal is indeed directly associated to the INCT InSAC thematic project ``National Institute of Science and Technology for Cooperative Autonomous Systems Applied to Security and Environment'' (Thematic FAPESP 2014/50851-0 and CNPq 465755/2014-3). Hence, this proposal also contributes to develop national competences, solve problems related to national security and environment, and open up new research directions and new international collaborations.