Stabilization and georegistration of aerial video over mountain terrain by means of lidar

Stabilization and georegistration of aerial video are important problems that can be difficult to solve, especially over mountainous regions. We introduce an approach that registers video frames to images predicted from a digital elevation model derived from LIDAR. It does not require reference imagery, because three-dimensional features such as rocks and vegetation can be predicted and registered with the corresponding features in the video imagery. The algorithm recovers the exterior and interior camera orientations with accuracy, and it accomplishes stabilization, georegistration and orthorectification of the video. This enables many applications, including fusion with satellite imagery, overlay of map graphics, and detection of moving objects. After describing the algorithm, we present results on several motion imagery datasets, including infrared and color video, and several types of elevation data.