Super-Resolution of Moving Vehicles Using an Affine Motion Model

This study addresses the problem of motion estimation and image registration for super-resolution (SR) algorithms. It appears that, simple assumptions such as planar motion are not always appropriate for dealing with real surveillance video of unconstrained moving objects. A new SR algorithm based on an affine motion model is developed and evaluated on a database, which contains videos of urban scene. Experiments show improved SR performances with respect to methods using only planar motion.