基于光度与点线特征融合的半直接单目视觉定位算法  

Semi-direct monocular visual localization algorithm based on photometry and point/line feature fusion

作　　者：彭清漪 夏林元[1] 吴东金[1] PENG Qingyi;XIA Linyuan;WU Dongjin(School of Geography and Planning,Sun Yat-Sen University,Guangzhou 510275,China)

机构地区：[1]中山大学地理科学与规划学院,广东广州510275

出　　处：《传感器与微系统》2020年第4期110-113,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(41704020);中央高校基本科研业务费专项资金资助项目(17lGPY43);国家重点研发计划资助项目(2017YFB0504103)。

摘　　要：针对室内环境中对定位精度和可用性不断提高的需求,基于视觉的同步定位与建图算法(SLAM)成为了室内定位领域的研究热点,但是视觉SLAM算法或因采用点作为特征导致其在低纹理环境鲁棒性较低,或因只采用光度计算而对环境过于敏感。由此,提出了基于光度和点线特征融合的半直接单目视觉定位算法,结合基于光度的直接法构建相对稠密的逆深度场景结构进行位姿优化,同时在基于点特征的算法上加入直线特征提高特征的丰富性和位姿估计的准确性。通过EuRo C数据集和实际场景实验表明:所提方法具有更高的位姿估计精度和鲁棒性。Aiming at the increasing demands for positioning precision and availability in indoor environments,simultaneous localization and mapping( SLAM) based on vision has become a hot field in indoor positioning research,but the visual SLAM algorithm may be less robust in low-texture environments due to the use of points as features,or it may be too sensitive to the environment because it uses only photometric calculations. A semi-direct monocular visual positioning algorithm based on the fusion of photometric and point/line features is proposed,combined with the direct method based on photometry to construct a relatively dense inverse depth scene graph to optimize poses,and adds straight line features to the point-based algorithm to improve feature richness and pose estimation accuracy. Experiments on the EuRo C dataset and actual scene show that the proposed method has higher pose estimation precision and robustness.

关 键 词：点线特征 光度 半直接单目视觉定位 

分 类 号：P232[天文地球—摄影测量与遥感]