Research On License Plate Recognition In Complex Environment

License Plate Recognition System,as an important part of intelligent transportation system,is becoming more and more important in recent years.In practical situations,the images collected by monitors are not always the pictures in perfect environment.It makes the whole process of plate recognition more difficult,or even causes the failure of the entire image recognition.The demand of the real-time and accuracy of license plate system is higher compared to other image recognition systems.At present,it is still not perfect for license plate recognition technology in complex environment although both domestic and foreign scholars have done a lot of researches.In real situations,the reason of the poor quality of taken images mostly attributes to the foggy environment.This paper takes an important processing for sprayed images,does lots of researches for various parts of the license plate recognition and achieves the license plate recognition in complex environment.In the part of image defogging,the reason of degraded image with poor quality is firstly analyzed.Then,the dark channel prior method is adopted to do the image defogging and the disadvantages of algorithm is also analyzed and improved.The method that uses surrounded pixel dot transmittance to optimize central dot transmittance is proposed in eliminating the Halo phenomenon.The tolerance mechanism is proposed in improving color inconsistency and the transmittance is improved by using tolerance parameters and the most optimal value is obtained according to the experiments.In the part of license plate positioning,the initial location and fine location are processed for the license plate.In the stage of initial location,according to the value in RGB component of the background color in license plate,the pixel statistic method for license plate is initially located by combining prior knowledge of the license plate texture.In the stage of fine location,in order to reduce the complexity of further processing,the tilt correction is conducted for license plate firstly and the method combining the mathematical morphology with pixel statistics is adopted to realize fine location and eliminate the interference of up and down the border,rivet and so on.In the stage of license plate character segmentation,the vertical projection is adopted to achieve the license plate character segmentation through combining license plate inherent feature.In the stage of character recognition,according to the BP neural network,the method that the characters in different locations is directly recognized for different networks has been put forward.The parameters in various networks are designed and the optimal value isdesigned by adjusting the parameter value for partial parameters and analyzing the influence of parameters for network performance.The method to join momentum factor is proposed to adjust the weight and prevent the excessively used phenomenon of the weight and adjust momentum factor to make the network performance the most optimal.Finally,the effectiveness of license image recognition in complex environment is verified through lots of simulation experiments.