Combining hough transform and contour algorithm for detecting vehicles' license-plates

Abstract

Vehicle license plate (VLP) recognition is an interesting problem that has attracted many computer vision research groups. One of the most important and difficult tasks of this problem is VLP detecting. It is not only used in VLP recognition systems but also useful for many traffic management systems. Our method is used for a VLP recognition system that deals with Vietnamese VLPs and it can also be applied to other types of VLPs with minor changes. There are various approaches to this problem, such as texture-based, morphology-based and boundary line-based. In this paper, we present the boundary line-based method that optimizes speed and accuracy by combining the Hough transform and contour algorithm. The enhancement of applying the Hough transform to contour images is the much improved speed of the algorithm. In addition, the algorithm can be used on VLP images that have been taken from various distances and have inclined angles between /spl plusmn/30/spl deg/ from the camera. Especially, it can detect plates in images which have more than one VLP. The algorithm was evaluated in two image sets with an accuracy of about 99%.