Research On Angle-free License Plate Recognition

With the rapid increase of the number of motor vehicles,traffic management departments are facing increasing pressure on vehicle supervision.The development of intelligent transportation system is of great significance for alleviating the pressure of traffic management and improving transportation efficiency.Stable and efficient license plate recognition is the basis of the intelligent transportation system,which plays an important role in improving the efficiency of access control,violation accountability,road inspection and other tasks,it is currently a research hotspot in the field of intelligent transportation system at home and abroad.The license plate recognition system is mainly located in the road crossings,entrance gates and other positions,with more restrictions on the camera attitude,stable illumination and single background,so the license plate recognition is relatively easy to achieve.However,the current mainstream methods have poor adaptability to license plate tilt and distortion,and the accuracy of license plate recognition under difficult angles cannot meet the actual requirements.Although the research of automatic license plate recognition has made great progress and has been widely used,stable and efficient license plate recognition methods under complex environment still need to be researched.Pointing at the above problems,this study aims to improve the accuracy of anglefree license plate recognition and explore a complete set of angle-free license plate recognition solutions.The main research contents and results are as follows:(1)Angle-free license plate feature dataset.This study specifically investigated the layout standards of the current commercial automatic license plate recognition system,and summarized the characteristics of the open source license plate datasets at home and abroad.It was found that the current datasets did not contain the pose information between camera and license plate,such as height and angle.Aiming at the deficiency of existing datasets,I collected the dataset which included the pose information between camera and license plate.After dataset annotation,data enlargement and data organization,an angle free license plate recognition dataset containing the relative pose information of license plate camera is constructed,which provides high quality data basis for model training and method evaluation of angle free license plate recognition.(2)A method for vehicle license plate location and correction based on depth feature.In view of the current surveillance video data in the license plate image and camera smaller angle between the main axis,feature extraction method based on convolution neural network is studied and developed in the traditional convolution neural network as the skeleton,fusion with distortion correction ability of spatial transformation network license location correction methods,precise positioning license plate location and complete the correction,obtaining high quality license plate image,reduce the difficulty of license plate character recognition.、(3)End-to-end license plate recognition method.Consider about the characteristics of angle-free license plate recognition image,an end-to-end license plate recognition method is proposed.Further through the induction and sorting of target detection,character recognition and other methods,explore the license plate recognition methods suitable for this study,and put forward a complete set of anglefree license plate recognition solution.