The Improvement Research Of Vehicle License Plate Recognition System Based On Bayonet

Entering the 21 st century,with the continuous development of social modernization,the number of car ownership around the world continues to grow,and Intelligent Transportation System(ITS)has gradually become a key step to solve the problem of urban traffic management.Automatic License Plate Recognition(LPR)can effectively register and manage vehicle information.It is an important research field of ITS.LPR is widely used in actual road traffic management such as speed detecting,illegal snapping,urban traffic monitoring,and vehicle travel time statistics.In the areas of public security monitoring in communities,schools,hospitals,and parking lots,LPR has also been widely used.This paper makes a comprehensive understanding of the main algorithms and related theories of LPR in recent years and systematically analyzes the difficulties of LPR's technology.The core algorithms in the LPR technology: license plate detection,character segmentation and recognition are deeply studied.The upgrade and optimization were also implemented on the hardware,and the development of license plate recognition system on the PC side was completed.The main work of this paper is as follows:(1)In the aspect of license plate detection,an AdaBoost cascade classifier combined with a maximum stable extreme regions(MSER)detection algorithm was proposed.The license plate could be located accurately through sequential processing.This method effectively overcomes the unfavorable factors such as illumination,tilt,external character interference and so on.The effect of counterfeiting is remarkable.The experimental collected 6816 pieces of license plate-containing sample data under different conditions of place,light,weather for the off-line testing.6784 pieces of license plates were detected accurately.The average accuracy of the license plate positioning module was increased to 99.53%.(2)In terms of license plate character segmentation,license plate character segmentation preprocessing is first performed,which mainly includes tilt correction and inverse color transformation,and a horizontal correction method based on MSER license plate character detection is proposed.The experimental results show that when the license plate tilt angle is large and the image quality is poor,the proposed method is better than the Hough transformmethod to detect tilt angle.The correction accuracy of 6784 plates with different degrees of tilt has been 99.75%.After segmentation preprocessing,the license plate character is segmented accurately based on the initial segmentation of the sliding template combined with the adaptive projection character boundary adjustment.The experiment of integrated license plate character segmentation module,the accuracy of the character segmentation module which including segmentation preprocessing reached 97.32%.(3)In the aspect of character recognition,the features of the Histogram of Oriented Gradient(HOG)were selected as feature extraction of the characters,and the Support Vector Machine(SVM)is selected to perform character recognition finalilly.The experimental comparison is carried out by the classifier which is trained by different features;the experimental result showed that the HOG feature can extract the edge shape features of the license plate character more effectively than the LBP feature through statistics misidentified license plate characters;and finally the selected character feature is the HOG feature.After optimization and expansion of the license plate character library and feature parameters,the accuracy of character recognition is further improved.(4)In the design of overall system,the construction of a license plate recognition system combined with industrial cameras and automatic triggering devices based on QT platform was implemented.A good user interaction interface was designed through the QT platform.In combination with the industrial camera and the infrared triggering device,a whole set of license plate recognition system based on the bayonet was realized.The complete process of the license plate recognition can be realized for both off-line pictures and real-time video captured vehicle.In the actual scene,experiments were carried out on the license plate recognition of the 2,500 license plates captured in real time,among which 2406 license plates were identified correctly,and the comprehensive recognition accuracy reached 96.25%,and it's time consumption is meet the real-life requirements.