Real-Time Video Stabilization for Micro-Air Vehicles

Real-time stabilization of videos aquired by Micro Air Vehicles (MAVs) is a challenging problem due to unconstrained platform motion and the quality of video frames. As the potential of video exploitation using MAV platforms is being realized, real-time stabilization of MAV videos for presentation to warriors as well as for subsequent exploitation has become a key technology enabler. In this Phase I effort, SET Associates proposes to develop a real-time video stabilization algorithm appropriate for EO video cameras on MAVs. Our approach fuses two complimentary approaches: one based on optical flow and the other on tracking horizons. Both approaches are inspired by biological systems. The optical flow based approach yields real-time stabilization but suffers from lack of robustness due to errors in computing optical flow. The horizon-based approach is more accurate, but can only estimate pitch and roll. Our fusion approach achieves more robust stabilization through the use of an extended Kalman filter that incorporates error covariances of the optical flow and horizon tracking algorithms. We will demonstrate evaluate the performance of the fused algorithm on a variety of MAV video clips.