Research And Application On Key Technology Of License Plate’s Recognition

With the rapid economic development and people’s living standards’ improvement, the popularity of the vehicle becomes an inevitable trend. Intelligent TransportationSystems have also emerged which effectively alleviated the traffic control and vehiclemanagement problems. LPR (License Plate Recognition) is an important component ofIntelligent Transportation System, which integrates computer vision, image processing,pattern recognition to monitor each of a motor vehicle on the road all day. LPR mainlyhas four parts of plate location, character segmentation, tilt correction and characterrecognition.In the license plate location, the paper adequately consider the color characteristicsof licenses in our country, achieve color segmentation in the HSV color model space andthen the plate’s region is located by the texture of vehicle plate. The algorithm solved theadverse impact of shooting angle and complex background which happens in practicalengineering application of the license plate locating, the location rate of this algorithm isproved at93.1%. In order to improve the result of character segmentation andrecognition, license plate picture after location should be corrected in the license platerecognition system. A tilt correction method by minimizing variance of coordinates of theprojection points is proposed. In the horizontal tilt correction, edge detection will beadded to the algorithm. To increase the accuracy of the vertical tilt correction, increasesthe level of interference removing part of the frame plate after completing the correction.Both precision and algorithm complexity of the method have great advantages inengineering application. The vehicle license’s segmentation is realized by a newsegmentation method combines with connected component method and the binaryvertical projection of license plate image. Then, single character is recognized by BPneural network, this paper designed BP network of characters, BP network of letters, BPnetwork of digitals and letters, the rates of recognition are91.3%%,94.6%and93.8%.Applied the algorithms for testing,260samples were collected and249can be identified,the recognition rate is about90.4%.