Research And Imnlementation Of License Plate Detection And Recognition System In Complex Scenarios

The large amount of data resources in the era of big data,and the substantial increase in computing power of hardware computing resources have led to the.rapid development of artificial intelligence neural network technology.In recent years,urban brains and smart security have received the attention of many researchers.However,there are few studies on license plate detection and recognition in complex security scenarios.Most of the papers are limited to license plate detection and recognition in a relatively single scene.Most of the implemented systems only support scene detection with only one license plate and limited types of license plate recognition.The recognition effect on small and two-row license plate is poor,and there is not enough support for license plate attributes.Aiming at the disadvantages of the traditional license plate detection and recognition system,this paper optimizes the license plate detection and license plate recognition algorithm to adapt to more complex application scenarios,increase the license plate detection rate in the pictures containing multiple license plates,and improve the recognition algorithm on the two-row license plate.At the same time,a license plate correction algorithm is proposed to correct the deformation of the license plate in a large-angle scene,which makes the input of the recognition algorithm more standardized and further improves the recognition algorithm.In addition,according to the efficiency problem in the actual license plate detection and recognition system,a algorithm for determining the license plate quality is proposed,which improves the system’s operating efficiency and increased the speed by more than 10 times.For different application scenarios,two systems are implemented in this article,namely the license plate detection and recognition system based on limited scenes of video stream and the picture-based license plate detection and recognition platform system.The license plate detection in this paper optimizes the FPN structure based on RetinaNet,and proposes a Fuse-FPN structure.The license plate correction module proposes two structures with and without residuals,which are respectively applied in different scenarios.The license plate recognition algorithm proposes a two-branch optimization method and a TR-Attention optimization method,which significantly improves the recognition accuracy of two-row license plates.The license plate quality judgment put forward the PQA(Plate Quality Judgment Algorithm)to predict the license plate quality score.