Automatic license plate detection in hazardous condition

Abstract

Automatic detection of license plate (LP) is to localize license plate region from an image without human involvement. So far a number of methods have already been introduced for automatic license plate detection (ALPD), but most of them do not consider various hazardous image conditions that exist in many real driving situations. Hazardous condition means an image can be a ected by rainy or foggy weather, may have low contrast environments (such as, indoor and night), can be blurred, having other objects in the background and may have horizontally tilted LP area. All these issues create challenges in developing e ective ALPD method. In this thesis, we propose a new ALPD method which can e ectively detect LP area from image in hazardous conditions. In the proposed ALPD, several innovative steps are introduced for handling the inherited issues of hazardous conditions. For rain removal, a novel method is applied that uses frequency domain mask to lter rain streaks from an image. The proposed rain removal technique performs better than the existing single-image-based rain removal approach (Kang et al. 2012). A new contrast enhancement method with a statistical binarization approach is introduced in the proposed ALPD for handling low contrast indoor, night, blur and foggy images. For correcting tilted LP, a novel Radon transform based tilt correction method is applied. To lter non-LP areas, a few unique approaches are used which are based on image entropy and average horizontal counting. This new ALPD method is tested on 850 car images having di erent hazardous conditions and achieves satisfactory results in LP detection. We also compare the performance of the proposed ALPD method with two existing state-of-the-art methods (Wen et al. 2011 and Hasan et al. 2013). The proposed ALPD method shows best performance among them in terms of detection rate and average running time.