Research On The Algorithm Of License Plate Recognition In Complex Scenes

With the development of economy science and technology,the number of vehicles has been gradually increasing which has caused the traffic problem to become more and more prominent.As a result,intelligent transportation technology has gradually entered into the spotlight.License plate recognition is a key technology in the field of highway traffic automation.It has an important application value in public safety,traffic management,customs,military and other departments.And it has become a hot research topic.The background of the vehicle is complicated.At the same time,the license plate image is affected by the weather conditions and shooting conditions.The recognition effect of common license plate recognition algorithms in such complicated scenes is not satisfactory.This study makes some improvements on the basis of the existing license plate recognition algorithms and proposes a license plate recognition algorithm suitable for complex scenes.The main work is as follows:(1)An appropriate illumination compensation algorithm is used for harsh weather conditions or night scene images.According to different color of vehicle license plates,the characters are different in colors.The research focuses mainly on extracting and analyzing the color characteristics of the license plate area,detecting areas that may contain license plates by using the HSV color space with high accuracy and stability.According to the character edge and color rule of license plate,continuous multi-peak feature detection is performed to screen out the coincidence area.Finally,the authenticity of the license plate is judged by the SVM model with excellent classification performance.Experiments show that the proposed algorithm not only applies to complex scenes,but also improves the detection accuracy.(2)Based on the existing license plate segmentation algorithm,combined with the collected license plate data,the segmentation algorithm is implemented.Aiming at the different quality images,a color channel-based binarization is proposed.In addition,after the correction of the tilt of the license plate and the determination of the upper and lower boundaries of the license plate character,in the character segmentation stage,the characters' relative distance and aspect ratio are fully utilized to solve the problem that the license plate characters are stuck and the Chinese characters are not connected.(3)The convolution neural network Lenet-5 is introduced into the license plate character recognition.According to the particularity of license plate characters in our country,some modifications were made to the model structure and two networks were trained for the identification of numbers/letters and Chinese characters respectively.Experiments show that the recognition performance of the model is better than the traditional Lenet-5 model in recognizing license plate characters.