License plate automatic detection and recognition using level sets and neural networks

In this paper, we propose a method for automatic license plate detection and recognition in the city of Abu Dhabi. The proposed method starts by segmenting moving vehicles using background subtraction. Segmented vehicles are tracked using a color-based particle-filtering technique until the vehicle is in position for a high resolution image to be taken by a still camera. The license plate is detected by converting the image into the LAB color space and using level set methods to locate its contour. Regularity and size are used to filter erroneous blobs. Geometric features are extracted from the blobs of license plate numbers and are passed to trained neural networks for classification. A model of the proposed system is built and its operations verified. Results show the proposed system's ability to determine a vehicles authorization status from the recognition of the license plate class or color as well as its number.