Algorithm of car license plates location based on multi-feature fusion

Videos in an intersection were taken synchronously by panoramic camera and close-range camera, the image was sampled by close-range camera in calibrated location of panoramic camera, car license plate is located in this close-range image. Aiming to some features of car license plate, edges are detected by using improved color Sobel operator, and the binary image is processed by adopting the method of dilation, fill and erosion in mathematical morphology, then several candidate license plate regions are obtained. By using some eigenvalues, such as areas, the ratio of height to width and vertical projection, the car license plate can be located from these candidate regions by adopting fuzzy neural networks. Experimental results demonstrate that this method has a high locating rate.