Robust Video Georegistration in the Presence of Significant Appearance Changes

Video information can provide an inexpensive source of information about the world. For many applications such as surveillance, situation awareness and navigation, the utility of this video information is increased if we are able to assign precise geocoordinates to the pixels in the video acquired from an airborne platform. Many video-capture platforms have physical sensors which can give an approximate relationship between the video and the world. For example, unmanned aerial vehicles (UAV) can transmit to a ground control station, together with the video, some telemetry information given in terms of the location and attitude of the platform relative to a world coordinate system, focal length and pose of the camera. While this telemetry, or Engineering Support Data (ESD), is very useful in giving a preliminary alignment of the video to the world, it may have limited accuracy due to a variety of factors: very precise inertial navigation system (INS) are generally expensive; the weight of the inertial measurement unit (IMU), which for some implementations prescribes the accuracy, can be limited by the maximum payload of the platform; a small jitter in the cameras can translate to significant ground errors for oblique or high altitude plat-forms. Video processing has the potential of improving on the precision of the alignment beyond what can be obtained with physical sensors alone.