Locating license tags using edges

Regular-polygon detectors are used in many applications including, license plate recognition, driver assistance, microscopy, aerial imagery etc. Given the enormous number of methods available for detecting polygons from an image, many of these approaches are robust, but are unsuitable for real time systems. In this paper we propose a fast and efficient method for detecting and localizing the regular polygons in an image. In particular, the proposed detector is used for detecting license plate in a License Plate Recognition (LPR) system. However, the detector can be used in other applications as well. Our main observation is that the edges separated by a regular angular spacing forming a polygon, can be easily distinguished from other edges in the image. The edges are grouped perceptually and localized based on their orientation and position. We propose a simple polygonality score that measures the difference between the product of magnitude and relevancy of the opposite sides of a polygon. Finally, a fast non-maximal suppression on the detected boxes is achieved by clustering the results based on their intersection over union (IOU). The proposed system runs at above 30 fps with an accuracy of 93%.