Research On The Algorithm Of Vehicle License Plate Recognition System Based On Computer Vision

With the improvement of people’s living standard, the vehicle starts to become a household necessity. As a result, the number of vehicles is increasing rapidly and a large number of problems appear, such as traffic congestion, traffic accidents and vehicle theft. The intelligent transportation system(ITS) arises at this point in order to solve these problems. Vehicle license plate recognition(VLPR) system as a key part of the intelligent transportation system, has been widely applied in high-speed road charges, parking fees, residential vehicle entry and exit management, illegal surveillance and other fields. In this paper, vehicle license plate recognition algorithm is the main research content. It consists of three modules: license plate localization algorithm, license plate characters segmentation algorithm and license plate characters recognition algorithm.(1) License plate localization algorithm. According to the videos and images collected by the cameras in different situations, this paper studies a license plate localization algorithm based on machine learning combined with the texture features. Firstly, This algorithm uses the machine learning method to realize the rapid detection of vehicles. Then, relatively stable plate texture features are choosed as the key factor for license plate localization on the vehicles. This paper also proposes some effective ways to remove the noise edges and constructs a classification model for distinglishing the plate area and non-plate area, finding the real plate area. This algorithm effectively improves the license plate localization accuracy and robustness.(2) License plate characters segmentation algorithm. According to the license plate features, this paper studies a license plate characters segmentation algorithm based on projection features combined with dynamic templates. This algorithm takes some ways, including license plate accurate positioning, arrangement of the characters and projection analysis, to realize the pre-segmentation of the license plate. Based on the dynamic templates, this paper uses the sliding window optimization method to find the segmentation positions of license plate characters, realizing the accurate segmentation of the license plate characters.(3) License plate characters recognition algorithm. This paper designs the effective features of the simplifying histograms of oriented gradients and the grid. According to these effective features, this paper makes the classification and discrimination of the characters for the first time. Against similar characters recognition, this paper increases the feature dimension and designs a multi-level support vector machine(SVM) classification model for the classification and discrimination of the characters for the second time. This algorithm effectively improves the character recognition accuracy and robustness.This paper presents a license plate recognition system based on VS2013 developing platform. A large number of images taken under different scenarios are used for license plate recognition algorithm validation and testing. The proposed algorithm experimentally has a series of advantages, such as high accuracy, robustness and so on.