The Research And Implementation Of License Plate Localization And Character Segmentation Based On Embedded System

License plate recognition (LPR) is the technique to extract the character information from the vehicle image and confirm the identity of the vehicle through using computer vision, image processing, pattern recognition, and so on. Because of its highly flexibility, embedded LPR (E-LPR) becomes an important study field in the Intelligent Transportation System (ITS), and has great apply value in the common security, traffic management, flow observation and electronic charge system.E-LPR system consists of embedded hardware system and software algorithms. This paper conducted in-depth research on the above parts, and ultimately achieved a complete E-LPR system.In the software part, the existing license plate location and character segmentation algorithm was ported to DSP platform and optimized on it. The optimization was carried out in three levels:algorithm level, C language level and assembly level. In the algorithm level, the original program was optimized for better recognition rate and running speed. The optimized floating-point algorithms were then converted to fixed-point algorithms without declining the recognition accuracy. In the C language level, the critical loops of the program were optimized using software pipelining technique, which is base on the pipeline and parallel processing unit of C64 DSP. Finally, in assembly level, some time-critical functions were rewritten in assembly language to take advantage of the advanced DSP instruction set and other resources. After the above three levels of optimization, this license plate location and segmentation algorithm can run real-timely on DSP platform.For hardware, a full-functioned E-LPR hardware platform was designed. This hardware circuit use TMS320DM642 as core processor, including 4 video captures, Ethernet communication module, and external memory module, etc. A board support package (BSP) was also developed for this circuit, providing driver functions including board initialization, video capture driver, FLASH operation library, I2C driver, etc. At last the automatic boot process was designed using a second boot mode. The hardware circuit, the BSP and the boot process together constitute a complete hardware platform.The software and hardware was integrated based on the DSP/BIOS real-time kernel, to form a complete E-LPR system. This system uses C/S mode with a PC as client, DSP platform as server and Ethernet as the communication method. The client submits images to the DSP system. After the DSP system recognizing the license plates, the result will be transmitted back to the client. The server system uses multi-threaded mode, and can parallel the image receiving, processing and sending work.This E-LPR system can process one image within 40ms, which meets the real-time requirements. Also, it has no limits on the location of plate in images and the image size, which gives it a better recognition rate and robustness.