Research On Self-positioning Technology Of UAV Based On Vision And Inertial Measurement Unit

There are a large number of application requirements for realizing the positioning of the UAV(Unmanned Aerial Vehicle)based on vision and IMU(Inertial Measurement Unit).The visual information can be used to estimate the bias of the IMU and to correct the trajectory drift,at the same time inertial information not only can be used to locate the UAV in the case of missing vision features,but also to recover the scale information of monocular vision.Therefore,this paper builds a visual and inertial positioning system to realize self-positioning of UAV.The main research work can be summarized as follows:(1)In the visual information processing,the data association of adjacent image frames is established based on the feature point method and the optical flow method,respectively.Firstly,the grid-based processing method is used to eliminate the feature redundancy of traditional feature operators.To solve the mismatching problem in the feature point method,this paper proposes Cross-match filter and two-points RANSAC(Random Sample Consensus)combination method to eliminate the mismatching.For the mistracking problem of optical flow method,this paper proposes to use NCC(Normalized Cross Correlation)and eight-point RANSAC combination method to remove the outer points.Finally,some experiments are designed to verify that the proposed algorithm has accuracy and real-time performance in eliminating mismatch.(2)In the IMU data processing,in order to avoid IMU data re-integration when updating initial value during nonlinear optimization,the pre-integration technique is used to preprocess the IMU data.Firstly,the IMU kinematics model which considering noise and bias is established,and then the error propagation in the pre-integration process is derived to the process of pre-integration calculation,Finally,processing the noise and bias realize IMU pre-integration.(3)To initialize the system,through combining visual information,the gyroscope’s bias is first estimated,then the scale factor,velocity,and gravity are roughly estimated.Finally,the velocity and scale factor are accurately solved by optimizing the gravity direction.(4)In the fusion of visual information and inertial information,a tightly coupled nonlinear optimization model framework based on sliding window is adopted.Firstly,the optimization method is introduced to solve the system state and the residual term,and its Jacobian matrix in the optimization process is derived.Then the old frame information is marginalized to ensure continuous consistency of the positioning trajectory.Finally,the optimized position information is sent to the ground control system through UDP.(5)In order to verify the effectiveness of the visual and inertial positioning system,experiments were conducted in the public dataset and the actual scene respectively.Firstly,the feasibility of the system is verified on the public dataset.Then,in the actual scene,multiple trajectory experiments are carried out in the indoor environment and the outdoor environment respectively.The positioning error is less than 1.5% in the indoor scene,and the average error between outdoor and GPS trajectories is less than 1m.The effectiveness and robustness of the proposed algorithm are verified.