Tracking in Persistent Wide-Area Motion Imagery

Persistent wide-area motion imagery is an emerging technology that provides continuous, unified, city-wide view using an airborne multi-camera array system. Analyzing these images for all moving objects is challenging for vision algorithms due to high data volume of several terabytes per hour, georegistration and stabilization errors, low frame rate, low spatial resolution, strong parallax, clutter from distractors, large number of similar objects, significant appearance change of objects and occlusions. Existing approaches for vision-based tracking of vehicles are not directly applicable since they rely upon high-resolution foreground objects and higher frame rates. We have developed an interactive tracking algorithm that addresses many of these challenges and can incorporate street maps or lidar data when available. The tracking algorithm performance is demonstrated on data obtained from a wide-area imaging platform.