Design And Implementation Of License Plate Location And Character Recognition System Based On ZedBoard

With the rapid development of Chinese economy, ITS (Intelligent Transportation System) has been a major trend in the development of future traffic monitoring system. As LPR (License Plate Recognition) technology is one of the core technologies of ITS, the research and development of license plate recognition system has extremely important significance and application value for Chinese transportation management.Through consulting a great number of relevant references, the majority of license plate recognition systems are based on PC software. The system interfaces and functionalities are fixed, resulting in poor scalability. LPR system still needs improvement, especially at adaptability, accuracy and speed. Therefore, this thesis puts forward a method to design and implement a license plate location and character recognition system based on ZedBoard development board. This system takes full advantage of dual-core ARM Cortex-A9 processor and FPGA on ZedBoard. Using hardware-software co-design to realize system functions improves the processing speed greatly and satisfies system’s demand better.In this thesis, the LPR system is composed of three modules:license plate location module, license plate character segmentation module and license plate character recognition module. By studying the key technologies of the above three modules, this thesis puts forward effective solutions. Firstly, in terms of license plate location module, this thesis puts forward a method combining the color and texture message. This module employs Verilog HDL language to extract licence plate candidates, detect license plate horizontal tilt angle and vertical tilt angle, which greatly improves the speed and accuracy of license plate location module. Secondly, in terms of the license plate character segmentation module, this thesis puts forward a method using modified projection algorithm. This method removes license plate frame and rivets firstly, and through modified projection algorithm to segment characters. This method can solve segmentation errors caused by adhesion characters and fractured characters, and improve characters segmentation accuracy. Finally, in terms of the license plate character recognition module, this thesis puts forward a method combining feature extraction and BP neural network algorithm. This module recognize characters through normalize characters, extracts coarse mesh features and periphery features of Chinese character, coarse mesh features and stroke density features of letter and figure character, and then design BP neural network.Moreover, this thesis achieves cross-platform transplantation of the Linux operating system, OpenCV and QT library, and also designs the test control module. As tests show that the location accuracy of the system is more than 90%, recognition accuracy is more than 89%, and the average running time of the system is about 2s, which meets the system requirements and achieves the expected design purpose.