Algorithm for automatically producing layered sprites by detecting camera movement

Abstract

We clarify the relationship between motion vectors (two parameters) and all camera motion, namely scaling, rotation and translation. All camera motions are described as Hermart transform coefficients. After all scaling and rotation factors of camera motion are removed, translation factors represent the depth of real three dimensional world. So we proposed a new sprite model, which is consist of planes vertical towards the depth direction. Using this sprite model, any background reflecting any camera motion creates sprites. Our simulation results show clear sprites and a clear synthesized image. The correlation between the original image and the synthesized image is more than 0.7. That indicates that the created sprites are suitable for prediction.