Research On Algorithm Of License Plate Recognition In Complex Environment

With the continuous improvement of China's economy,the number of cars is increasing,so it is necessary to apply the advanced nature and effectiveness of the Intelligent Traffic System(ITS)to manage the vehicles.The ITS can efficiently allocate resources,improve road load capacity and traffic management efficiency,so as to release the pressure of urban traffic congestion.The Vehicle License Plate Recognition System(LPRS)is a major component of the ITS,and has become a research hotspot in recent years.At present,the LPRS has begun to be widely used in traffic fields,such as intelligent toll collection?monitoring of illegal vehicles and electronic police.However,the license plate recognition products currently put into practical applications have the disadvantages of lack of universality and robustness,that is,in a complex environment,such as the license plate is reflective due to light,the license plate color is defaced,the surface of the license plate is dusty,the foreign matter on the license plate is blocked,the illumination is uneven at night,and even a plurality of vehicle license plates exist in a traffic image,so the efficiency and accuracy of license plate recognition will be seriously reduced.In view of the above situation,the technical research on license plate recognition in complex environments still has a long way to go.In view of the above problems of license plate location,character segmentation and recognition in the research of license plate recognition in complex environment,this thesis proposes some effective improvement measures.1.In the license plate location algorithm,a method of generating a gray image based on license plate color features and binarizing is adopted,in order to eliminate the interference points around the license plate,the binary image generated based on HSI space and the binary image mentioned above are logically processed.Then the initial license plate location is carried out according to the texture and structure features of the license plate,and the Hu moment and PHOG features are extracted from the candidate regions and fused.Finally,the precise license plate location is carried out using PSO optimized SVM.The experimental results show that the method has a success rate of 97% in a complex environment.2.In the character segmentation algorithm,aiming at the common problems of horizontal tilt and inaccurate angle detection based on Hough transform,the combination of Hough transform and rotating horizontal projection overcomes the shortcoming of single method,the algorithm based on jump method and vertical projection method to remove the border not only saves time greatly,but also ensures the removal result.A segmentation algorithm based on the character structural feature and traversing vertical projection is proposed for character segmentation,the algorithm is easy to implement and has less computation.According to the experimental results,it can be seen that the algorithm can avoid the influence of the separation between the second character and the third character.At the same time,it can correctly cut the license plate characters under the conditions of fracture?fading?foreign object occlusion?illumination interference and so on.3.In the character recognition algorithm,In view of the problem of LeNet-5 based on traditional CNN structure in recognition when the number sample is insufficient,this thesis improves the network from the activation function and structure,simplifies the structure by removing the C5 layer and increases the dropout strategy at the full connection layer.The improved network achieves the improvement of the correct rate and efficiency of character recognition,the recognition accuracy rate is up to 93% and has high robustness.