Research On License Plate Recognition Method Of Vehicle Bayonet Scenario

To effectively improve the quality of video image data in public security and fully tap into the value of video images,the Ministry of Public Security has focused on the application and control demands for key individuals in recent years.Key efforts have been made to manage the basic information of video surveillance,vehicle bayonet,and micro-bayonet capture devices,as well as face and vehicle capture image data.As the only identification for motor vehicles,efficient license plate recognition can accelerate the improvement of the basic quality of video image data in public security and enhance the level of intelligent application of public security video images.Currently,license plate recognition under normal conditions has achieved good results.However,the recognition accuracy of existing algorithms needs to be improved for problems such as tilted,rotation,and blurring of images caused by shooting angles and environments for bayonet license plate images.Therefore,this paper conducts research on license plate issues in the context of road checkpoints.The specific research content is as follows:First,an attention mechanism based license plate recognition scheme is proposed for complex vehicle bayonet license plate recognition scenarios.The scheme adopts a encoder decoder structure,in which the encoder uses a residual network with a feature pyramid model to extract image depth features,and optimizes the network structure for low resolution license plate images.The decoder adopts attention mechanism without mask,which uses the attention mechanism to learn the global features of the image and strengthen the correspondence between the model about the license plate character features and the label,while removing the mask mechanism to convert the model autoregressive output into parallel output to improve the recognition efficiency.The experimental results show that the algorithm in this paper has higher accuracy in recognizing the license plate images at card gates,which proves the effectiveness of the algorithm.Secondly,in response to the problem of low recognition accuracy of blurry license plates,on the basis of the vehicle bayonet license plate scene,in-depth research is conducted on the recognition of blurry text,and a scheme for recognizing severely blurry text based on weakly supervised graph-text information joint recovery is proposed.The scheme uses visual and text modal information to reconstruct text images,with the visual modality used to restore the overall features of the text and the text modality used to strengthen the restoration of local detail features of the text.At the same time,the model uses a weakly supervised mechanism to obtain the input image’s text token as the text label through prior models,the model improves the recovery of text image details by text modal reinforcement structure without text labels.Constructing a benign feedback model where the visual modality forwards the text modality.Experimental results show that the reconstructed text images obtained using this scheme have higher recognition accuracy.