摘要
为了实时获取城市楼宇三维模型,以及估计楼宇遮蔽对目标探测的影响,构建了基于直接法和特征法的融合实时双目同步定位和地图构建(SLAM)系统。该系统主要由前端追踪、局部建图和闭环检测3个部分组成。前端追踪在无人机做有速运动时,采用直接法估计当前帧位姿;若直接法失败或无人机近似静止时,采用基于极线和三角测量精度2种约束条件的特征法估计当前帧位姿。局部建图和闭环检测部分采用随机蕨(Fern)模型构建查询数据库,从而简化了查询数据库的构建流程。在EuRoC开源数据集下进行了系统有效性验证,并与快速特征点提取描述(ORB)-SLAM2开源系统进行了试验对比分析。试验结果表明,该系统减少了追踪耗时,提高了定位精度,增强了系统实时性与鲁棒性。
In order to obtain 3D model of in urban buildings in real time and estimate the influence of building shelter on target detection,the real-time binocular simultaneous localization and mapping(SLAM)system based on direct method and feature-based method fusion is constructed.The system is mainly consists of the front-end tracking,the local mapping and the loop monitoring.In the part of front-end tracking,when the UAV’s speed is not zero,the direct method is adopted to estimate current frame’s pose.While the direct method failed or the UAV is almost stationary,the feature-based method based on the two constraints of epipolar and triangulation accuracy is adopted to estimate the current frame’s pose.In the part of local mapping and loop monitoring,the random fern(Fern)model is adopted to establish the query database,which simplifies the process of query database construction.The effectiveness of the system is tested under the EuRoC open source dataset and compared with the oriented fast and rotated brief(ORB)-SLAM2 open source system.The experimental results show that this system can reduce the tracking time,improve the positioning accuracy,and enhance the realtime and the robustness.
作者
李英琦
杨荣
陶章志
LI Yingqi;YANG Rong;TAO Zhangzhi(The 28th Research Institute of China Electronics Technology Group Corporation,Nanjing 210023,China)
出处
《指挥信息系统与技术》
2022年第1期51-57,共7页
Command Information System and Technology
基金
工业与信息化部国防科工局“十三五”预研课题资助项目。
关键词
直接法
特征法
双目视觉
Fern数据库
direct method
feature-based method
stereo vision
Fern database
