FPGA Design And Implementation Of ORB Image Feature Extraction Algorithm

With the continuous development of computer technology and the continuous enrichment of image processing theories,real-time detection of feature points to achieve vision goals has increasingly become a research hotspot in the field of computer vision.The image feature point algorithm is more commonly used in the registration and fusion of remote sensing images in the field of surveying and mapping geographic information,and it is of practical significance to realize the real-time extraction of feature points.Scholars in the field of image processing at home and abroad have proposed many corresponding solutions to common problems encountered in practice,focusing on how to improve the robustness and real-time performance of the algorithm.In recent years,with the emergence and rapid development of many processor platforms,especially the emergence of FPGA(Field Programmable Gate Array)field programmable gate array technology characterized by repeatable configurability,high frequency,and parallel processing,the data throughput of the image processing system is greatly reduced.There is a significant gain in processing power.Therefore,more and more research projects choose to put the image algorithm system on the FPGA platform for hardware acceleration.This paper takes ORB(Oriented FAST and Rotated BRIEF)image feature extraction algorithm as the research object,uses software and hardware collaboration technology as a means,and uses the high-level synthesis tool Vivado HLS provided by Xilinx to complete the RTL-level(Register Transfer Leve)IP(Intellectual Property)core packaging design of the ORB image feature extraction algorithm.Without losing the positioning accuracy of feature points,to improve the real-time performance of image processing,the processing platform based on the ZYNQ architecture realizes the hardware acceleration of the ORB image feature extraction system.The main work of this paper is as follows:(1)Systematically introduce the most important image feature detection key technologies and related theories,and reproduce these algorithms,and give the corresponding image processing results,so as to have an intuitive understanding of the related theories.(2)Analyze the principle of the ORB image feature point extraction algorithm,then compare other classic algorithms for feature point extraction,implement software implementation of these algorithms,give their detection results in the same scene,and then compare their Algorithm performance is evaluated.(3)Systematically introduction Xilinx HLS(High Level Synthesis)development platform,with the help of the unique stream data format and optimization instructions in HLS,write ORB algorithm implementation code that meets HLS parallel acceleration and pipelining specifications in HLS development environment,and perform IP that meets AXI4(Advanced e Xtensible Interface 4)transmission specifications The core package is packaged for use on the Vivado platform.(4)In the Vivado electronic design automation environment,configure the CMOS(Complementary Metal-Oxide Semiconductor)image acquisition module,VDMA(AXI Video Direct Memory Access)storage access module,timing configuration module and HDMI(High Definition Multimedia Interface)format video output display module accordingly,and add the ORB image processing IP core to complete the hardware platform construction.Finally,complete the design of the entire ORB feature real-time extraction system in the SDK,and burn it to the ZYNQ7020 development board for functional verification.