The License Plate Recognition For Traffic Crossroads Based On Feature Mapping Learning

With the rapid development of cities,the number of automobiles has increased dramatically,which not only brings convenience to people’s lives,but also causes serious traffic problems.Nowadays,in most urban surveillance scenarios,the recognition of license plate still relies heavily on manual recognition method,which requires a lot of manpower and material resources.It is prone to errors,and unable to work twenty-four hours,which unable to meet the needs of rapid development of cities.Existing license plate recognition methods can only accurately recognize license plates under simple test environment and good image acquisition conditions,they can not be applied in complex environment.Therefore,there is an urgent need for an automatic license plate recognition method which can accurately identify the license plate in complex environment such as urban traffic crossroads.Based on the method of deep learning,this paper designs a license plate recognition algorithm based on feature domain transformation,which realizes the accurate recognition of complex and changeable license plates.Then this paper improves a license plate detection algorithm,and proposes an automatic license plate recognition method for the complex traffic crossroads by combining the detection algorithm with the recognition algorithm,which can accurately detect and recognize the complex license plate,it meets the actual license plate recognition task requirements.This paper mainly includes the following two aspects:Firstly,aiming at the problems in the existing license plate recognition algorithms,this paper proposes a fully-convolutional license plate recognition method.This method adopts a sequential way to avoid error caused by character segmentation.By building excellent feature extraction module,efficient license plate sequence predictor and decoder to complete the one-shot recognition of license plate algorithm.Compared with other sequential integration recognition algorithms,this algorithm has higher recognition accuracy and speed.At the same time,this paper analyzes the existing problems of the license plate dataset,and made real and generated two datasets which has certain theoretical research value.Secondly,aiming at the problems of unstable recognition effect with deformation license plate in the fully-convolutional algorithm,we propose a feature domain transformation network to make complex image transformed into a simple and uniform feature,which can effectively reduce the interference caused by complex scenes.Then,this paper introduces the recognition loss function,a license plate recognition algorithm based on feature transformation is designed by combining image preprocessing algorithm with license plate recognition algorithm organically.This method enhances the performance of the whole algorithm and it is end-to-end trainable.Experimental results show that this method is effective and can recognize license plates accurately under complex traffic crossroads environment.The recognition accuracy reaches 96.49%.In conclusion,the method of license plate recognition based on deep learning proposed in this paper can accurately recognize license plate automatically.It achieves good performance which meet the actual needs of automatic license plate recognition under complex traffic crossroads environment.It has certain theoretical value and practical significance.