Research On The Vehicle License Plate Recognition System For Car Video Streaming

License plate recognition technology is the core part of the intelligent traffic system and is a typical combination of computer vision,image processing and pattern recognition technologies.At present,almost all the license plate recognition systems are fixed,based on this,this paper conducts a comprehensive study on the vehicle license plate recognition system for car video streaming.In view of the movable characteristics of the system,this paper first carries out the restoration processing on the blurred video frames,and then carries out license plate positioning and identification.(1)Blurred restoration section.Based on the spectrum analysis of the degraded image,Radon transform is used to obtain the tilt angle of the vehicle motion blur image,and the first-order differential autocorrelation function of the image is used to obtain the motion blur length of the image.Since the image restoration will cause the ringing effect,and some clear images do not need to be restored,based on the estimation of the length of the motion blur,this paper first makes a blurred decision on the image and then decides whether it needs to restore the image.Through experimental comparison,a better Wiener filtering method is used in the image restoration part.(2)License plate positioning section.Based on the full study of the classical license plate location algorithm,a license plate location algorithm that combines HSV color space and texture features is presented.The algorithm fuses the license plate candidate regions extracted based on the HSV color space and the license plate candidate regions extracted based on the edge detection,highlights the license plate area information..This algorithm makes up for the insufficiency of a single license plate location algorithm.It can accurately locate the license plate area when the camera is stationary and moving.(3)Character segmentation section.This paper presents an algorithm for segmentation of connected regions combined with license plate features.The algorithm is based on the inclination correction of the license plate image,binary cutting and morphological processing are performed on the binary license plate image,and the character is accurately cut in combination with the prior knowledge of the license plate character,which effectively solves the problem that characters cannot be accurately segmented due to character adhesion.(4)Character recognition section.This paper presents an improved convolutional neural network algorithm to identify the license plate characters.For the differences between Chinese characters and alphanumeric structures,two convolutional neural networks are designed to identify the two separately,thereby improving the recognition accuracy of license plate characters.