Robust Real-Time SFM for SMAV

We shall investigate and implement robust Real-Time SFM for SMAV application scenarios. Automatic pose determination and extraction of 3D structure of the environment are critical to autonomous navigation and obstacle avoidance of SMAVs in constrained and crowded environments. Extracting such state information under various constraints (low quality video, sensor payload limitations, minimal onboard processing capabilities, sudden and abrupt motions changes etc.) pose a significant challenge and require development of innovative and robust algorithms. We will 1) Develop and implement SFM algorithms with robustness at each processing stage to ensure reliable performance with low quality video input, and to detect degenerate or near degenerate cases to avoid erroneous estimation; 2) Perform extensive feasibility study of our proposed algorithm using data with progressive levels of fidelity including a) Simulated data with ground truth information b) Motion-captured data and c) Data collected from real SMAVs equipped with miniaturized cameras; and 3) Systematically evaluate the performance of our SFM algorithm for reconstruction accuracy at different flying patterns in various environments, and different depth ranges that can be recovered by our algorithm. The objective is to develop a practical and field deployable system that will provide navigation and obstacle detection capabilities to the SMAV platforms.