License Plate Recognition Algorithm Under Special Circumstances

In the era of rapid growth in the number of cars,traffic problems are becoming increasingly prominent.In order to improve the level of intelligent transportation,intelligent transportation technology emerges as the times require.License plate recognition is an important field of intelligent transportation,which has produced many important research results in the field of license plate recognition.In the traditional license plate recognition system,there are license plate location,license plate correction,license plate character segmentation,license plate character recognition steps.The errors will be accumulated in different modules,so that the final recognition rate is discounted.Especially in some special environment with insufficient illumination,the influence of error is more serious.Therefore,it is still of practical significance and market value to study the license plate location technology in special environment.This paper has the following points in the specific work:First of all,under the premise of ensuring the texture characteristics of the image,the image is enhanced by using the homomorphic filtering technology in the frequency domain.Several common filters of homomorphic filtering have too many parameters,the calculation is complex,and the parameter value selection is difficult.Aiming at these problems,the traditional Butterworth filter is improved,which is similar to Butterworth filter Instead of Butterworth filter,the function of waver changes from five parameters to two,and the complexity of function is also reduced.The experiment shows that the efficiency of image operation in enhancement operation is improved,and the illumination effect of image is well compensated.Secondly,in the process of license plate location,aiming at the common color and texture features of blue and yellow license plates,a license plate location algorithm based on the two channel difference between pixels is designed.In RGB space,histogram analysis is carried out for two kinds of license plates.In order to highlight the yellow and blue background colors of license plates,the middle value of the three components of red,green and blue is set to zero,and all pixels are binarized to get the pre-processing image.The bottom color and characters of license plates in the pre-processing image are different,so the jump point image of license plate characters is extracted,and then the license plate position is determined by projection method Therefore,this algorithm can locate two kinds of color license plate at the same time,which combines the characteristics of license plate color and character texture,does not depend on the edge information of license plate,and improves the efficiency of location significantly.Then,on the basis of location independent of license plate frame,aiming at the problem that license plate tilt correction mostly depends on license plate frame,the algorithm of image low rank texture invariance is introduced.In the tilt correction experiment of license plate,the effect is good.Only according to the minimum low rank of non tilt license plate,the correction of borderless license plate is realized.In the process of character segmentation,a character template is inserted to solve the problem of inaccurate segmentation and character breaking in traditional projection algorithm,which reduces the probability of character breaking and makes the segmented characters more complete.Finally,on the basis of the original Le net-5,the network structure is improved to be a double-layer Le net-5 neural network.According to the character characteristics of the license plate,the parameters and excitation functions of the model are adjusted,and the network is optimized and trained.The results show that the improved Le net-5 recognition efficiency is better than the original Le net-5 model.Compared with the template matching and BP neural network,it has obvious advantages.