A robust tracking algorithm for super-resolution reconstruction of vehicle license plates

Abstract

We propose a novel, very robust method for tracking a vehicle license plate in a sequence of low-resolution frames acquired by a video surveillance camera in order to reconstruct the license plate view in a super-resolution image. The tracking method is able to follow the license plate corners position with sub-pixel resolution and to compensate for small non translational spatial movements of the target during the motion by adopting a perspective transformation. In the reconstruction of the target each frame is perspectively transformed, aligned, cropped, de-convolved and interpolated to higher resolution. Eventually the data are combined into a super-resolution image.