Hardware design for registration of aerial video imagery

Aerial videos are used in mapping and surveillance applications, especially for updating satellite imagery. For the purpose of video mosaicing, we have used the Levenberg-Marquardt least squares minimization algorithm. Since this is a computationally intensive process, the bottleneck part of the algorithm i.e. sum of squared differences has been moved to hardware. This paper presents a cost effective, optimized design for registration of aerial video imagery. The emphasis is on the design of data path, such that the amount of hardware is minimized while keeping mosaicing error measure below a certain threshold. Furthermore, pixel caching is used to reduce the data transfer to and from memory. Finally, some of the benchmarks and simulation results are discussed.