Research On License Plate Detection And Recognition Algorithm In Complex Scenes

In recent years,driverless,intelligent transportation systems and smart cities have become hot topics in the field of transportation.License plate detection and recognition are widely used in parking lot toll collection systems,highway checkpoints,violation supervision,and other scenarios.The algorithm based on depth learning is more stable than traditional methods.However,it is difficult to detect and recognize license plates in complex scenes,including weather effects,remote license plates,and tilted license plates.To this end,this paper studies license plate detection and recognition in a variety of complex scenes through deep learning:First,a license plate region detection algorithm based on improved YOLOX-s is designed.To improve the accuracy of license plate detection in complex scenes such as light interference of license plates and remote license plates,the MSR image enhancement method is used for image adaptive enhancement during preprocessing to make the image texture clearer.New prediction branches are constructed in the feature extraction network to reduce the missed detection of small license plates.In addition,a coordinate attention mechanism is added to the network to strengthen the weight of useful information along different dimensions,Finally,the CIo U loss function is used to improve the convergence ability of the network.The experimental results show that the average accuracy of the improved algorithm is 98.25%,and the detection speed is34 FPS.Second,design an end-to-end SG-CRNN license plate character recognition algorithm.To solve the problem that tilted license plates are challenging to recognize,the space transformation network module is used to correct after detecting the license plate area.In addition,the depth bidirectional GRU and CTC methods are used in the algorithm to extract the context information of the input character feature sequence and output the character results with maximum probability.The experimental results show that the recognition accuracy of the proposed algorithm in the complex background is97.8%,and the average recognition speed is 28.6ms.Third,design an intelligent license plate recognition system.The intelligent license plate recognition system is composed of the license plate detection algorithm and the character recognition algorithm.The GUI interface that can directly recognize the license plate image is designed.The system performance is tested on the data set.The experiment shows that the recognition accuracy of the license plate recognition system developed in this paper reaches 97.4% in complex scenes,and the average recognition time is 42.6ms.Figure [33] Table [13] Reference [72]...