Research On License Plate Recognition Algorithm With Large Angle Tilting

License plate recognition is a very important basic function in intelligent transportation system,which is widely used.The current license plate recognition algorithm in the standard scene is relatively mature,and the recognition rate of the mainstream license plate recognition system in the market is 'very good under the standard scene.However,in many application scenarios,such as parking lot entrances and exits,the vehicle traffic direction is different,the license plate are often large inclined Angle in the image.The large inclined angle of the license plate has a great influence on the license plate detection,which will cause the license plate to be missed,at the same time,it will also cause character sticking and character extrusion deformation,and reduce the recognition accuracy of the license plate characters.In view of the above problems,this paper has carried out in-depth research on the detection,correction and character recognition algorithms of large angle slanted license plates.The main research contents include:First,the accurate license plate detection algorithm based on East net is constructed.Firstly,the necessity of applying accurate detection(detecting the four vertices of the license plate character area)for large angle tilting license plates is analyzed.Secondly,the East net structure based on the output of four vertex coordinates is realized.The main process includes:firstly,the four-channels distance output from the pixel point to four sides of the rectangle is modified to the eight-channels distance output to the four vertices,secondly,the IOU(cross-over ratio)is calculated using the minimum circumscribed rectangle of the quadrilateral surrounded by four vertices.Finally,this net model is used for accurate detection of the license plate character area.The experimental results show that,compared with the traditional algorithm,on the basis of ensuring the performance of the general license plate detection,the proposed algorithm can significantly improve the detection effect of large-angle tilting license plate,and the accuracy rate and the recall rate are more than ninety-seven percent.Secondly,the recognition algorithm of license plate characters based on two-step tilting correction is proposed.Firstly,the four vertex coordinates of the obtained license plate area are used for rotation transformation and perspective transformation,and the license plate image with a large Angle inclined is corrected and horizontally stretched to an appropriate size,and it overcomes the problems of character adhesion and character extrusion deformation caused by large Angle tilt;Then,the first Chinese character of the license plate is extracted from the corrected license plate character image,and the Chinese character is recognized by the LeNet-5 net to improve the recognition rate of Chinese characters.Finally,the CRNN net is used to end-to-end character recognition of the last few license plate characters,which avoids the error caused by the traditional character segmentation process and can recognize license plate characters of any length.The experimental results show that the two-step tilt correction can improve the recognition rate by nearly ten percentage points compared with the CRNN net,and the first Chinese recognition net can further increase the recognition rate by fifteen percentage points to Ninety-six point four percent.