Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (13): 257-264.DOI: 10.3778/j.issn.1002-8331.2012-0028

• Engineering and Applications • Previous Articles     Next Articles

Indoor Active SLAM Based on Fisher Information

XI Yonghui, HU Shiqiang   

  1. School of Aeronautics and Astronautics, Shanghai Jiao Tong University, Shanghai 200240, China
  • Online:2022-07-01 Published:2022-07-01

基于Fisher信息的室内主动SLAM

席永辉,胡士强   

  1. 上海交通大学 航空航天学院,上海 200240

Abstract: Aiming at the problem that traditional visual simultaneous localization and mapping(VSLAM) algorithm is prone to be invalid due to missing features in indoor weak texture scene, an active SLAM algorithm based on maximum Fisher information for tilt control is proposed. The method is extended on the classical ORB-SLAM2 framework, adding Fisher information field construction module and tilt control module. Firstly, the three-dimensional space is divided into several voxels during visual tracking, and the Fisher information of each voxel is updated according to the spatial distribution of features to complete the construction of Fisher information field. Secondly, when the image is acquired by the camera encounters the situation of missing features, the voxel nearest to the optical center of the camera is found, and the direction with the largest Fisher information of this voxel is considered to be the optimal observation direction of the camera. Finally, the deflection angle of the camera after coordinate transformation is calculated, and the camera is rotated to the optimal observation direction through the airborne tilt to obtain the features in scene again, which makes the algorithm realize autonomous relocation after losing features. The improved algorithm is applied to the simulation platform of quadrotor UAV and the results show that the algorithm proposed can still estimate the pose of UAV accurately in real time when the traditional algorithm fails, which improves the robustness of the system.

Key words: weak texture, pose estimation, Fisher information

摘要: 针对传统的视觉同步定位与地图创建(visual simultaneous localization and mapping,VSLAM)算法在室内弱纹理场景中容易因为特征缺失而定位失败的问题,提出了一种基于最大Fisher信息量云台控制的主动SLAM算法。该方法在经典的ORB-SLAM2框架上进行扩展,增加了Fisher信息场构建模块与云台控制模块。在视觉跟踪的同时,将三维空间划分成若干个体素,根据特征点的空间位置分布更新每个体素的Fisher信息,完成Fisher信息场的构建;当相机获取的图像遇到特征缺失的情况,先找到离相机光心欧式距离最近的体素,以该体素Fisher信息量最大的方向作为相机此时的最优观测方向;计算出坐标变换后相机的偏转角度,通过机载云台实现相机转动到最优观测方向,重新获得场景特征,使得算法在丢失特征之后能够实现自主重定位。将改进后的算法运用到四旋翼无人机仿真平台,结果表明在传统算法失效的情况下,所提算法仍能实时准确地估计无人机位姿,提高了系统的鲁棒性。

关键词: 弱纹理, 位姿估计, Fisher信息