基于Kinect2.0的三维视觉同步定位与地图构建  

Three-dimensional visual SLAM based on Kinect 2.0

作　　者：王龙辉 杨光[1] 尹芳 丑武胜[1] 

机构地区：[1]北京航空航天大学机械工程及自动化学院,北京100191

出　　处：《中国体视学与图像分析》2017年第3期276-285,共10页Chinese Journal of Stereology and Image Analysis

基　　金：国家自然科学基金重点项目(No.61633002)

摘　　要：本文研究了近几年迅速发展的以深度相机作为传感器的视觉SLAM(simultaneous localization and mapping)。相比于传统视觉SLAM方案中采用的K-d树或者K-means算法,本文提出一种基于K-means++算法的闭环检测方案,修正了系统的累计误差,提高了系统稳定性和定位精度。在系统前端,选用了基于ORB(Oriented FAST and Rotated BRIEF)特征的特征点法。后端包括位姿图优化和闭环检测,位姿图优化借助g2o(general graph optimization)通用求解器来实现,闭环检测采用了基于二维图像特征的词袋库模型。实验结果成功构建了清晰的三维环境点云地图并计算出精确的运动轨迹。This paper study the visual slam algorithm,which is developed rapidly with depth camera as sensor in recent years. Compared to the traditional visual SLAM scheme used in the K-d or K-means tree algorithm,we propose a closed-loop detection scheme based on K-means ＋＋ algorithm,which corrects the cumulative error of correction system and improves the system stability and accuracy. In the front-end of the system,we choose the point features-method based on ORB feature. The back-end consist of pose graph optimization and closed loop detection,the pose graph optimization is realized by g2o general solver,and closed-loop detection adopts the bag of words model based on two-dimensional image feature.And the experimental results succeeded in registering the three-dimensional environment point cloud map and get an accurate motion path.

关 键 词：SLAM ORB K-means++ 图优化 闭环检测 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置] TP391.41[自动化与计算机技术—控制科学与工程]