Feature Point Matching System Based On BRIEF Descriptor

In the intelligent age,the widespread use of embedded devices has changed our lives.With the rapid development of computer vision technology,it has achieved relevant research results in the fields of multimedia technology,medical image processing,industrial visual inspection,navigation and positioning.However,since computer vision algorithms are mostly computationally intensive algorithms,the computational complexity is high that makes it difficult to meet the requirements of high rate and real-time on traditional PC platforms.Therefore,porting these algorithms to embedded devices and building high-speed processing structures has become a breakthrough in the field of computer vision.In this paper,a feature point matching system based on the BRIEF descriptor is proposed.The system is implemented on the Xilinx ZYNQ SoC embedded platform.It optimizes and improves the SURF(Speeded-Up Robest Features)feature point detection circuit implemented by the research group,and implements the hardware acceleration on the FPGA(Field Programmable Gate Array)by the BRIEF(Binary Robust Independent Elementary Features)descriptor algorithm.This design proposes a novel parallel structure and a 3-frame processing mode for the SURF feature point detection circuit and the BRIEF description circuit,so that it can meet the need of current high-speed device for real-time processing of continuous video streams.In this paper,the descriptor matching circuit is also designed to complete the BRIEF descriptor matching of adjacent frames of the video stream,and the coordinates of the matching feature points in the picture frame are obtained in real time.After integrating the feature point matching system based on the BRIEF descriptor,the AXI4 communication protocol master-slave interface is programmed for the system to enable data exchange between the system and the external DDR under software control.In this paper,the feature point matching system based on BRIEF descriptor works at 100 MHz clock frequency.For VGA continuous video stream with resolution of 640 × 480,the maximum processing speed is 292 fps(frame per second);The whole system achieved a low power consumption of 2.105 W.By comparing the results of the standard open-SURF and OpenCV software,the feature point matching system based on the BRIEF descriptor in this design achieves more than 93% matching accuracy in various scenarios.In this paper,the feature point matching system based on the BRIEF descriptor is implemented on the ZC706 development board of Xilinx ZYNQ-7000 series.The main research contents include the following four parts: hardware outline design of the descriptor matching system,introduction and optimization of SURF feature point detection circuit,the hardware acceleration design of BRIEF descriptor and the design of the descriptor matching circuit.The descriptor matching system designed in this paper realizes the detection and matching of SURF-BRIEF feature points at high speed,and can be used in various real-time applications such as visual odometer and image stitching.