Design Of License Plate Recognition System Based On RISC-V Core

In recent years,with the development of artificial intelligence and 5G and other technologies,the license plate recognition system has also transitioned from the less adaptable PC platform realization to the portable embedded terminal,putting forward higher requirements on the accuracy and speed of the existing license plate recognition system,especially under the conditions of complex detection environment,insufficient light and large angle tilt of license plates.At the same time license plate recognition system as an important subsystem of intelligent transportation system,camera equipment needs to collect and process massive data,and its security is facing serious threats and challenges.Therefore,it is of great practical significance to research and design a high performance embedded license plate recognition system with high speed,high accuracy and autonomous control and security.This paper proposes a scheme to build an on-chip license plate recognition SOC system on programmable logic devices using Hummingbird E203 RISC-V cores around these two aspects to realize license plate image acquisition,license plate recognition and license plate result display functions.Firstly,combining the composition of Hummingbird E203 kernel,the AXI protocol interface bus that can mount peripherals is pushed out and unused peripherals in the kernel are removed through the icb2 axi module in Hummingbird E203 kernel to realize the modification of the kernel and facilitate the subsequent building of license plate recognition system;Secondly,by analyzing the advantages and disadvantages of the algorithms related to license plate recognition and the difficulty of implementation on FPGA,feasible solutions are selected on the traditional algorithms related to license plate recognition,namely image pre-processing(color space conversion,median filtering,binarization of global threshold based on median filtering),license plate area extraction based on the combination of color features and vertical projection,character segmentation by projection method and character recognition based on improved template matching.Among them,character recognition is improved for the traditional template matching algorithm in the process of selecting the scheme.By combining numerical features and character grid features,the characters are segmented into 8*5 cells and the advantages of multiple feature recognition are combined to carry out recognition processing,which improves the accuracy and processing speed of character recognition;Finally,the hardware platform for license plate recognition is built on FPGA and designed with pipeline and parallel ideas to fully utilize the advantages of FPGA parallel computing to achieve hardware acceleration of the selected license plate recognition algorithm.Through the comparison of license plate recognition results under different conditions in real situations and the verification of power consumption,resource consumption,processing time and other aspects,the license plate recognition system implemented in this paper has the best recognition effect under ideal situations and the second best recognition effect under complex backgrounds and can achieve an accuracy rate of more than 90%,which can solve the problem of license plate recognition when interference occurs in the background to a certain extent.Meanwhile,this license plate recognition system basically meets the embedded platform design requirements in terms of recognition time and power consumption.