摘要
为提高机器人在复杂环境下的定位精度和实时性,提出一种基于Vision-IMU的同时定位与地图创建算法。算法主要分为4个线程进行：跟踪、特征提取、局部建图和闭环。跟踪线程使用IMU信息辅助直接法进行像素点匹配,而后最小化重投影误差和IMU项误差,得到初步的机器人位姿估计。特征提取线程完成关键帧的特征点提取和描述子计算。局部建图线程利用光束平差法得到更加精确的机器人位姿轨迹和环境信息。闭环线程检测闭环,并利用闭环优化提高机器人轨迹和环境信息的一致性。公开的Euroc数据集实验表明,算法能够处理光线变化环境,实时地得到更准确的机器人定位。特别是相机剧烈运动产生运动模糊时,算法依然保持较高的定位精度。
To improve the accuracy and real-time of the robot localization under complex environments,a simultaneous localization and mapping（ SLAM） algorithm based on vision and inertial measurement unit（ IMU） is proposed. The proposed algorithm is primarily comprised of four threads,which are tracking,feature extraction,local mapping and loop closing. The tracking thread matches pixels by a direct method with IMU information. Then,the initial robot pose is obtained by minimizing reprojection errors and IMU term errors.The feature extraction thread completes keypoints extraction of keyframes and calculates the corresponding descriptors. Local mapping thread obtains more accurate robot trajectory and environment information with bundle adjustment. Loop closing thread detects loop closure and executes loop closure optimization to improve the global consistency of the robot trajectory and environment information. The experiments on the open dataset Euroc show that the proposed algorithm can cover changing light of environments and achieves more precise robot localization real time. Especially,the proposed algorithm can maintain exacter localization when the violent movement of camera brings motion blur.
作者
姚二亮
张合新
张国良
徐慧
赵欣
Yao Erliang;Zhang Hexin;Zhang Guoliang;Xu Hui;Zhao Xin(Rocket Force University of Engineering, Xi＇an 710025, China;Chengdu University of Information Technology, Chengdu 610225, China)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2018年第4期230-238,共9页
Chinese Journal of Scientific Instrument
基金
青年科学基金(61503393)项目资助
关键词
同时定位与地图创建
视觉
惯性测量单元
光束平差法
simultaneous localization and mapping
vision
inertial measurement unit
bundle adjustment
