基于AGAST-BRIEF的图像匹配融合算法  

Image Matching Fusion Algorithm Based on AGAST-BRIEF

作　　者：谷学静 刘艳佳 周记帆 肖军发 GU Xuejing;LIU Yanjia;ZHOU Jifan;XIAO Junfa(College of Electrical Engineering,North China University of Science and Technology;Tangshan Digital Media Engineering Technology Research Center)

机构地区：[1]华北理工大学电气工程学院 [2]唐山市数字媒体工程技术研究中心

出　　处：《仪表技术与传感器》2024年第10期78-83,110,共7页Instrument Technique and Sensor

基　　金：唐山市沉浸式虚拟环境三维仿真基础创新团队项目(18130221A)。

摘　　要：针对传统算法AGAST在图像处于复杂场景条件下进行图像检测时,造成图像匹配结果精度低和实时性差等问题,提出一种基于AGAST-BRIEF的图像匹配融合算法。首先,利用高斯滤波对图像进行预处理,实现去除图像噪声干扰,保留图像边缘信息的效果,通过AGAST与尺度空间理论相结合进行特征提取;然后,使用BRIEF描述子对图像中的关键点进行描述,提高特征匹配效率;最后,采用FLANN算法得到初次匹配对,使用GMS与改进的RANSAC算法对初次匹配结果进行二次筛选,得到图像特征精匹配。实验结果表明:所提算法相较于SIFT、AKAZE、KAZE算法匹配准确率提高,匹配运行时间最短,具有良好的鲁棒性。Aiming at the problems of low accuracy and poor real-time performance caused by traditional AGAST algorithm in image detection under complex scene conditions,an image matching fusion algorithm based on AGAST-BRIEF algorithm was proposed.Firstly,the Gaussian filter was used to preprocess the image to achieve the effect of removing image noise interference and preserving image edge information.The feature extraction was carried out by combining AGAST and scale space theory.Then,BRIEF descriptors were used to describe the key points in the image to improve the efficiency of feature matching.Finally,FLANN algorithm was used to get the first matching pair,GMS and improved RANSAC algorithm were used to filter the first matching result,and the image features were matched accurately.The experimental results show that compared with SIFT,AKAZE and KAZE algorithms,the proposed algorithm has higher matching accuracy,shortest matching running time and good robustness.

关 键 词：图像匹配 快速最近邻逼近搜索函数库(FLANN) 自适应通用加速分割检测(AGAST) 二进制鲁棒独立基本特征(BRIEF) 基于网格的运动统计(GMS) 随机抽样一致算法(RANSAC) 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]