Research On Precise Positioning Of License Plate Based On Deep Learning

In License plate recognition technology is an important part of building intelligent traffic management system.The research on license plate location is a key step to realize license plate recognition.The traditional license plate location algorithm is greatly affected by the surrounding environment factors,such as the occlusion of the license plate,the dry weather,the license plate of different colors,etc.will have a great impact on the accuracy of the license plate location.This paper proposes a deep learning method to achieve accurate positioning of the license plate.Compared with the traditional license plate recognition algorithm,it has better robustness and accuracy in terms of illumination,occlusion,and non-licence area noise.The division and recognition of characters laid a good foundation.The in-depth study of traditional license plate location algorithms analyzes the flow of their algorithms in the license plate location and their advantages and disadvantages.The edge feature localization method is sensitive to noise in the image,especially when similar texture features or other interference factors in the image are encountered,the license plate has a high positioning error rate.The color feature localization method can simply describe the color distribution in the license plate image,which is not sensitive to the size and direction change of the image region,so the local features of the license plate cannot be extracted well.The multi-feature location method based on mathematical morphology is an improvement of the traditional mathematical morphology algorithm.Combined with the two license plate location algorithms of color feature and edge feature,it shows the multi-feature of the algorithm and effectively improves it.The speed and accuracy of the entire positioning.This paper focuses on the license plate location algorithm based on deep learning.Explain the various frameworks of deep learning and the environment we set up in the experiment.Finally,we design a three-level concatenated convolutional neural network architecture: the first layer is based on the target detection R-CNN algorithm for the detection of vehicles in the image,and the second The layer is theimage detected by the first layer as the data,and the rough detection of the license plate.The third layer is the coarse positioning of the license plate through the second layer as the data,and the key positioning of the license plate is performed.The overall network structure reflects the cascade.The idea of convolutional neural networks from coarse detection to fine detection.And through experiments,the cascading convolutional neural network is more accurate in the location of the license plate,and the detected recognition rate is higher than the traditional license plate location algorithm.