Moving Sensor Video Image Processing Enhanced with Elimination of Ego Motion by Global Registration and SIFT

This field of target tracking is relatively mature when the camera is stationary but the moving sensor poses uniquely challenging problems because relative to the camera, everything in the scene appears to be moving. This report will present a robust and efficient video stabilization algorithm based on the scale invariant feature transform (SIFT) algorithm and global registration. An important comparison will be made to KLT, a widely used feature based tracker, stabilization algorithms. We will demonstrate KLT's deficiencies and SIFT's advantages when confronted with limited features and 3D structures. Finally an implementation of the overall system with real data provided by UAV's from MRLets Technologies Inc will be described and analyzed.