单目视觉惯性融合方法在无人机位姿估计中的应用

Monocular Vision-inertial Fusion Approach for MAV State Estimation

为了提高四旋翼飞行器位姿估计精度,文章提出了一种多传感器融合算法。其设计了一个模块化的卡尔曼滤波框架,将视觉部分的初步位姿估计和基于惯性测量单元(IMU)的动力学模型相融合;提出了一种新的视觉惯性里程计方案,利用ARTag标签进行多标签融合的位姿估计,并将其结果作为算法的观测量;建立了基于IMU的动力学模型,将视觉结果加入状态向量不断进行迭代。由于IMU的频率高于视觉结果的频率,每当利用视觉结果进行更新时,选取最近的状态向量,这样在视觉结果异常时,该算法依然可以很好地工作。该算法的性能在实验中有很好的体现。

To improve the pose estimation accuracy of quadrotors MAV(mertial measurement unit), it presented a new algorithm of state estimation for quadrotors. In this paper, a modular Kalman filter framework was designed, which combines the initial pose estimation of the vision part and the IMU-based dynamic model, and this framework is a new visual inertial odometry. The ARTags were used to obtain pose estimation, and the result was used as the observation of the algorithm. The dynamic model based on IMU was established, and the visual result was added to the state vector for iterations. Since the frequency of IMU is higher than that of the visual result, the most recent state vector is selected when it updates. The algorithm still works well when the visual result is abnormal, and it shows a good performance in the experiment.