Maximum a posteriori image registration/motion estimation

A novel method is presented for on-line, recursive image registration/motion estimation. Comprising a Kalmari filter and an interlaced new image registration algorithm, the method enables the estimation of aircraft motion from a sequence of terrain images, acquired by an airborne, down-looking electro-optical sensor. Contrary to other methods of vehicle motion estimation that are based on the conventional mean-of-squared-differences (MSD) image registration algorithm, the new method utilizes the maximum a posteriori (MAP) estimation methodology to draw statistical information from the prediction level of the Kalman filter. The resulting image registration algorithm is thus rendered more accurate and robust with respect to loss of lock (measurement divergence). A small error statistical analysis shows that, under reasonable conditions, the new MAP algorithm is unbiased and efficient, and its estimation error covariance is smaller than that of the ordinary MSD algorithm. The superiority of the new MAP algorithm over the conventional MSD algorithm was substantiated via an extensive experimental investigation, using real aerial photographs.