Research For Vehicle License Plates Recognition System

License Plate Recognition (LPR) is one of the critical techniques for the intelligent transportation system. Vehicle license plates recognition system consist of four modules in general, those are: the collection of picture of vehicle, license plate location, character segmentation and character recognition. Focus on difficulties of LPR, the paper studies the key technologies of it, and finally settles down a series of effective algorithms used in LPR based on experimental results.Research for the collection of vehicle pictures. Some common approaches in the field of detecting and tracking moving targets including the approach based on optic flow, the approach based on frame subtraction and the approach based on background modeling are also introduced and compared with each other. Their advantage and disadvantage also with the field they fit are discussed in this paper. Through analysis and comparison of these methods, the background subtraction method is applied to detect moving vehicles. With the background modeling method, a system of detection and tracking moving vehicles for the static camera is realized there.In license plate location, it analyzes and compares many license plate location algorithms, such as edge detection method, artificial neural network method, gray method and method base on color, etc. In addition to, it present a license plate location algorithm based on vertical edge detection.In character segmentation, it analyzes and compares a lot of algorithms about license plate image binary and standardization and geometrical revising. And then a character segmentation method is used to segment character, which is base on projection pre-known knowledge of the size and the space between characters of vehicle license plates.In character recognition, it analyzes and compares many character recognition algorithms, such as artificial neural network algorithm. In addition to, it puts forward thought way of BP neural network of character recognition algorithm. Some methods are proposed to improve the convergent speed and precision of BP neural network.Experimental results show that these methods are reasonable, location rate of license plates , segmentation rate of characters, recognition rate of characters is more than 90%.