Research Of Foggy License Plate Recognition System Based On Improved Retinex Algorithm

The Intelligent Transportation System(ITS)has become the main means to solve traffic problems such as traffic pressure,road congestion and so on.License Plate Recognition System(LPRS),which is one of the core of ITS,is more and more involved in practical traffic management,and has achieved good results.However,due to uneven illumination,weather changes and other complex environments,the quality of license plate images obtained by the camera is not good,so the accuracy of license plate recognition is decreased.In this thesis,in view of the frequent occurrence of foggy weather in coastal cities,a license plate recognition system in foggy environment is studied and designed.After study and analysis,the thesis take the most common license plates with blue background and white characters as the research object.The thesis will study from four aspects,which including image defogging and enhancement,license plate location,character segmentation and character recognition.In order to improve the performance of license plate recognition system in foggy environment,this thesis studies and draws lessons from the existing scientific research experience,and puts forward its own solution according to the specific research background.First of all,enhancing the images collected by the camera in foggy weather.By understanding the basic theory of the Retinex algorithm,the Retinex algorithm based on local features is studied emphatically.It adopts a multi-scale Retinex algorithm based on HSV color space,and analyses the effects of image defogging according to subjective and objective evaluation.Secondly,locating the license plate position in the image.It converts RGB images to HSV color space,and locates the license plate roughly according to the color information.And the candidate regions are finely located according to the vertical projection method.Then,segmenting the license plate image.In this thesis,a segmentation method combining horizontal projection and template matching is proposed,which has a good segmentation effect for breaking or conglutinating characters.Finally,recognizing the single character.Several common feature extraction methods and character recognition methods are introduced firstly.The thesis uses particle swarm optimization(PSO)to optimize the parameters of support vector machine,and it is determined that the algorithm can be used to recognize thecharacters of license plate.In this thesis,the license plate recognition system in foggy environment based on the improved Retinex algorithm is studied,and the system flow is simulated in order to understand the license plate recognition technology better.After a large number of experiments,the method proposed in this thesis can meet the expected requirements and has a certain theoretical value and practical significance in intelligent transportation.