License Plate Recognition Technology Research Under Complicated Conditions

In today’s traffic management system, license plate recognition system is more and more highlight its importance.However, with the complicated traffic environment, the existing traditional license plate recognition gradually cannot satisfy the requirement of people.Especially for the night of speeding cars to capture,the vehicle images the Imaging equipment collected in the night are very poor,at the same time, the speed too fast lead to blurred images,the identification precision of license plate recognition system is greatly reduced.Therefore, it is necessary to carry out the study of night speeding car license plate recognition.Night speeding vehicle image degradation mainly causes low illumination resolution less, color information loss and motion blur, for these three kind of degraded image restoration problem has carried out the following research:For license plate color information loss and low illumination environment problem with large amounts of noise, put forward an improved adaptive wavelet threshold enhancement algorithm.First, use the Retinex theory to enhance the quality of the image under low illumination,then use the wavelet threshold function to restore the image.Aimed at the flaws of the traditional soft and hard threshold algorithm,propose an improved threshold function.the results show that the algorithm is effective on every scale of the wavelet.Aiming at the problem of speeding vehicles will appear ghosting fuzzy, proposed an improved iterative blind deconvolution restoration algorithm,ameliorate the traditional fuzzy algorithm to magnify the noise in the process of recovery. Put forward a blind deconvolution restoration method combining with noise estimation,the algorithm does not need to have prior knowledge of fuzzy parameters,combined with least squares estimate of the noise signal,it restored the blurred image point spread function(PSF) and fuzzy image itself at the same time.The experimental results show that the improved algorithm of processing results has a higher signal-to-noise ratio and can better remove the fuzzy.Aiming at the problem of color information loss.Because the traditional limitation of main license plate localization algorithm based on color information,put forward an improved Ada Boost machine learning algorithm.In view of the traditional rely only on the machine algorithm for license plate recognition accuracy rate lower defect, introduced the rough license plate location,the process can use the image edge information effectively.Then use the license plate image characteristics of Haar- like to train the Ada Boost classifier.Update sample weights through the iterations to get the weak classifier,the weak classifiers were combined to the strong classifier,use the strong classifier to sift the positive and negative samples.The experimental results show that,the join of the coarse position combined with the edge information greatly improved the efficiency and accuracy of algorithm,using the Ada Boost classifier can accurately extract the license plate area from vehicles image.After the Improved pretreatment and positioning process,the template matching using simple and efficient method of license plate character recognition. Experiments show that compared with ordinary pretreatment and positioning process,the process of license plate pretreatment process and recognition under low illumination and speeding combined with the wavelet enhancement denoising and Ada Boost algorithm,blind deconvolution can greatly increase the vehicle’s license plate number recognition rate.