Research On License Plate Recognition Based On Deep-Learning

License plate recognition technology already gains a high recognition rate under good external lighting conditions,but in foggy days,weak lighting,blurred or partially obscured characters and tilted license plates,the recognition effect of the existing algorithm is sometimes less than ideal and cannot meet application requirements of modern intelligent transportation system.In order to improve the accuracy and adaptability of license plate character recognition in complex scenes.This paper analyzes the characteristics and fuses improvements based on the research and analysis of the latest technologies related to deep learning,and applies them to the positioning and recognition of number plates,mainly doing the following work.(1)The YOLOv4 algorithm,which currently has superior target detection performance,is studied and analyzed,and based on its characteristics of being able to accurately and quickly detect small target objects,it is applied to the detection and localization of license plates.K-Means++clustering algorithm is used to cluster the license plate target frame before training data set,thus effectively improving the positioning effect.The experimental results show that the algorithm achieves 99.5% positioning accuracy in complex scenarios.(2)The tilt correction of license plates is investigated by using the NMS(Non Maximum Suppression)algorithm in combination with the perspective transformation to correct license plates tilted in different directions.Firstly,the algorithm preprocesses the license plate image,such as enhancement and denoising,then detects the edge,calculates the four vertex coordinates of the license plate,and deduces 8 values of the transformation matrix M in the perspective transformation through matrix operation to achieve the correction of inclined license plate.The experimental results show that the algorithm is less affected by the external environment and has good correction effect.(3)The existing algorithms for licence plate recognition based on deep learning are analyzed and their advantages are combined.A server layers convolutional neural network architecture is studied and designed,and the underfitting and overfitting problems in the training process are solved to achieve the recognition of localized and corrected license plates.The experimental results show that the algorithm can achieve good recognition results.