Combined Appearance-Motion Tracking for PGMTI (UAV-PGMTI)

SET Associates proposes a novel online appearance and motion encoded tracker for robust tracking of vehicles using a particle filter. The appearance of targets in the first few frames is modeled using a mixture of Gaussians and the tracking problem is posed as maximizing the posterior probability of finding this specific appearance using a particle filter. Robustness is achieved by combining motion and appearance in the particle filter; speed is achieved by using as few particles as are needed using the concept of asymptotic relative efficiency. The appearance model tracks the changes in vehicle appearance as it moves in the scene by updating the parameters of the mixtures of Gaussians model. We propose joint background/foreground modeling for platform motion cancellation, appearance models for false alarm suppression and real-time operation by using an efficient particle filter. A method for handling occlusion is suggested using robust statistics and predicted vehicle motion using the particle filter. We propose design concepts for representative UAVs and feasibility experiments to demonstrate detection and tracking. A Phase II plan will be formulated for implementing a real-time algorithm to demonstrate moving target tracking to support off-board target acquisition from a UAV for standoff Gunship aircraft.