基于RGB-D相机数据的SLAM算法研究  

Research on SLAM Algorithm Based on RGB-D Camera Data

作　　者：刘继忠[1] 王聪[1] 曾成 LIU Jizhong;WANG Cong;ZENG Cheng(Nanchang Key Laboratory of Medical and Technology Researh,Nanchang University,Nanchang Jiangxi 330031,China)

机构地区：[1]南昌大学机电工程学院,南昌大学南昌市医工结合技术研究重点实验室,江西南昌330031

出　　处：《传感技术学报》2021年第6期770-777,共8页Chinese Journal of Sensors and Actuators

基　　金：江西省重点研发计划项目(20202BBE53025,20202BBGL73057);南昌市高层次科技创新人才“双百计划”项目(2020131);南昌市医工结合技术研究重点实验室项目(2018-NCZDSY-002)。

摘　　要：提出了一种基于RGB-D相机数据的同时定位与地图构建(SLAM)算法系统,实现对RGB-D数据的快速和准确构建.首先在RGB图中提取较好鲁棒性的SURF特征,结合使用快速最近邻近似(FLANN)来完成特征点匹配的方式,再使用基于改进的最小距离与随机采样一致性(RANSAC)组合的方法剔除误匹配,然后使用PNP求解相邻帧间的相机位姿变换关系,后端的优化使用G2O(general graph optimization)来优化全局位姿,并使用回环检测消除累计误差.实验证明该方法具有有效性和可行性,能够迅速、准确地构建出三维稠密地图.A simultaneous localization and mapping(SLAM)algorithm system based on RGB-D camera data is proposed to achieve rapid and accurate construction of threedimensional dense map.Firstly,SURF features is extracted with good robustness from the RGB image,and Fast Library for Approximate Nearest Neighbors(FLANN)is used to complete the feature point matching method,and then the method based on improved minimum distance and random sampling consistency(RANSAC)are adopted to eliminate false matches,which applies PNP to solve the camera pose transformation relationship between adjacent frames.The backend optimization uses G2O(general graph optimization)to optimize the pose Graph,and uses loopback detection to eliminate the cumulative error.Experimental results prove that the proposed method is effective and feasible,which can quickly and accurately construct a threedimensional dense map.

关 键 词：同时定位与地图构建 G20优化 特征匹配 三维地图 RGB-D 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]