Camera motion estimation for object-based video coding

Abstract

This paper deals with motion estimation of an airborne camera and generation of 3D landscape models from image sequences for object-based video coding. While most camera motion estimators use a random sampling outlier elimination, the presented new approach uses a priori knowledge to simplify the complexity of the outlier elimination and to achieve real time coding. Test results with synthetic sequences show an estimation error of maximal one percent for the motion parameter which leads to a good quality for the generated 3D object models.