光学学报, 2018, 38 (7): 0715004, 网络出版: 2018-09-05
基于VI-SLAM的四旋翼自主飞行与三维稠密重构 下载: 1193次
Quadrotor Autonomous Flight and Three-Dimensional Dense Reconstruction Based on VI-SLAM
机器视觉 同时定位与地图构建 传感器融合 微型飞行器 三维稠密重构 machine vision simultaneous localization and mapping sensor fusion micro aerial vehicle three-dimensional dense reconstruction
摘要
提出全自主的微型飞行器,使用板载传感器实现三维的同时定位与稠密重构。在ORB-SLAM系统的基础上,基于扩展卡尔曼滤波器实现了视觉-惯导的传感器融合,提高了系统的稳健性和精度以满足微型飞行器自主飞行的要求。由于ORB-SLAM系统创建的稀疏的特征地图不能用于微型飞行器的避障和导航,使用双目摄像机提出了改进的构建地图的方法,由稀疏特征点地图扩展为稠密的八叉树地图。通过EuRoC数据集进行评估,可以验证本文算法较基于关键帧的视觉-惯导算法平均精度提升了1倍。将本文算法应用于所搭建的四旋翼自主飞行平台,仅依靠板载传感器和处理器,实现了全自主飞行与稠密地图构建,验证了本文算法的有效性和稳健性。
Abstract
We propose a fully autonomous micro aerial vehicle with onboard sensors to achieve simultaneous three-dimensional localization and dense reconstruction. Based on the ORB-SLAM system, a visual-inertial simultaneous localization and mapping system is proposed based on the extended Kalman filter, which improves the robustness and accuracy of the system to meet the requirements of micro aerial vehicle autonomous flight. Since sparse feature point maps created by the ORB-SLAM system can't be used for micro aerial vehicle obstacle avoidance and navigation, a stereo camera is used to propose an improved method of building maps from sparse maps to dense octree maps. The experiment evaluation with EuRoC dataset shows that the proposed algorithm improves the precision of open keyframe-based visual-inertial algorithm by one time. The proposed algorithm is applied to the quadrotor autonomous flight platform, and the fully autonomous flight and dense map construction is achieved by relying on on-board sensors and processors. The effectiveness and robustness of the proposed algorithm are verified.
林辉灿, 吕强, 卫恒, 王阳, 梁冰. 基于VI-SLAM的四旋翼自主飞行与三维稠密重构[J]. 光学学报, 2018, 38(7): 0715004. Huican Lin, Qiang Lü, Heng Wei, Yang Wang, Bing Liang. Quadrotor Autonomous Flight and Three-Dimensional Dense Reconstruction Based on VI-SLAM[J]. Acta Optica Sinica, 2018, 38(7): 0715004.