| As an important automatic identification technology, license plate recognition(LPR) system has been applied in numerous practical applications, includingintelligent transportation systems, traffic surveillance, tolls collection on highways,access control for security and so on. Generally speaking, according to its executioncourse, LPR can be separated into three steps: license plate location (LPL), charactersegmentation and character recognition. Since there are lots of diverse and uncertainconditions when capturing vehicle images in natural scenes, it is generally recognizedthat LPL is the most critical and challenging task in LPR.In this paper, we present a fast and efficient license plate location method incomplex environments. Based on the vertical gradient of the vehicle image, we firstselect the region of interest (ROI), which most possibly contains the plate area, andconcentrate only on ROI in the following process. Next we introduce an efficientalgorithm to enhance the gradient by considering both local intensity variance andlocal edge density, then remove most of redundant edges by a new approach, andfinally search and segment the plate by the revised mean shift algorithm and slidingan adaptive rectangle window in the residual edge image.Based on the experiments performed on348gray images in size of384×288,the location accuracy of the proposed method is95.7%, and meanwhile, its averagecomputational time of a single image is only7.7ms. Experimental results show thatthe new method possesses strong robustness to environmental noise and provides highaccuracy of real-time license plate location. |
