Research On License Plate Recognition Technology In Unrestricted Environment

License plate recognition system,as one of the key technologies of modern intelligent transportation system,has been applied in more and more different scenes.In addition,the existing license plate recognition method has a high recognition rate and can meet the requirements of the actual situation under the condition of better illumination conditions and relatively fixed vehicle image acquisition locations,such as highway checkpoints and toll stations.However,when the license plate recognition system is used in an unqualified environment,such as auditing,violation detection,and justice system,due to the camera to the vehicle position,Angle,light and image is the change of environmental background texture,a variety of characteristics of license plate image,such as geometry,gray and texture,there will be a different change.Therefore,in the application scenarios with unrestricted conditions,the system has poor adaptability to environmental changes,which leads to the reduction of LPR rate and fails to meet the application requirements of modern intelligent transportation system.In order to improve the adaptability of the license plate recognition system under unrestricted conditions,this thesis makes improvements according to the main steps of license plate recognition,which are license plate image preprocessing,license plate location and license plate recognition.Firstly,in vehicle image preprocessing phase,aimed at the phenomenon of particle image caused by the atomizing air,put forward a kind of based on dark channel prior to fog optimization algorithm combined with wavelet transform,first using the dark channel prior principle to thick fog,collected images by wavelet transform respectively after the source image to fog and since the high frequency part of image and low frequency part and mutual confluence,thus the collected image preprocessing,prevent the air atomization in the impact of vehicle image segmentation.In the phase of vehicle image segmentation,a vehicle image segmentation method suitable for non-uniform illumination is proposed.Secondly,according to the characteristics of gray-level image imaging under non-uniform illumination,gray-level crest and gray-level trough were selected,and the selected crest and trough values were taken as the limiting conditions to limit the scope of the initial population of genetic algorithm.Then,the improved double-threshold segmentation method was adopted to segment the vehicle image under non-uniform illumination.Next,the segmented vehicle area is extracted and the license plate location method based on HSV is adopted to locate the license plate part.Finally,the Tesseract engine is used to recognize the license plate characters in the image.Method is to use the Tesseract’s supporting training tool--jTessBoxEditor,finishing the license plate training set,after finishing the training set to make use of a Tesseract engine training,and generate new dictionary.Then Tesseract-ocr recognizes license plate characters in an image through a newly generated dictionary,test results show that adopting the new dictionary after training was carried out on the license plate recognition accuracy compared with the original it has a high recognition rate.To sum up,this thesis proposes a license plate recognition method under unrestricted conditions.In the image preprocessing stage,the collected images are defogged and optimized to improve the clarity of it.After that,the vehicle image under the condition of uneven illumination is segmented by an improved dual-threshold segmentation method based on genetic algorithm,and then the vehicle part in the segmented image is extracted.The license plate location based on HSV is adopted to process the vehicle part,and the image containing the license plate area is obtained.Finally,the new dictionary is trained by the Tesseract engine,and the processed license plate images are imported into tesseract-ocr for recognition.Due to the performance of Tesseract-OCR,the license plate characters can be well recognized.Through the test of the test set of license plate data,it can be concluded that the license plate recognition method proposed in this thesis can carry out license plate recognition under the environment of unqualified conditions and can continuously improve the recognition accuracy by training,which has certain application prospect and theoretical value.