Research On Indoor Positioning Of UAV Based On Visual Inertial Odometry

Indoor positioning technology has become a hot research topic in the field of Unmanned Aerial Vehicle(UAV).In low speed condition,the angular velocity and acceleration of the component itself measured by Inertial Measurement Unit(IMU)has obvious drift,which causes the obtained position and attitude of UAV has deviation from the actual one,while visual sensor performs well.On the other hand,in high speed condition visual sensor has motion blurring due to fast motion,and IMU does not have such a problem.In the thesis,IMU will be used for fast motion measurement and visual sensor will be adopted in slow motion condition.For the "outliers" produced by sensors,the measurement noise covariance matrix will be modeled based on the inverse Wishart(IW)distribution,which makes the measured noise follow the t-distribution.Then the heavy-tailed characteristic of the outliers noise can be simulated.The main work of this paper is demonstrated as follows:Firstly,an attitude measurement method for multi-rotor system will be proposed based on IW distribution.Then the simulation will be performed with input from the collected data of IMU gyro and accelerometer.For comparation,the simulation with noise “outliers” as input will also be carried out.The results based on Extended Kalman Filter(EKF)and the proposed attitude method will be compared to show the robustness to heavy-tailed noise.Additionally,the visual odometer and the IMU will be fused in a loosely coupled manner.The IMU data will be used as input to predict the states of UAV,and then the output of the visual odometry will be adopted to correct such states.In order to deal with the heavy-tailed noise of visual output data,a visual inertial integrated navigation method for heavy-tailed noise will be proposed by using IW distribution.Based on the dataset collected on-board a Micro Aerial Vehicle(MAV),the experiment result will be used to verify the effectiveness of the fusion algorithm and the single visual odometry.Then the outliers of the measurement noise will be introduced,and the fusion result will be compared with the result of EKF fusion framework to show the robustness of the proposed fusion algorithm for heavy-tailed noise.