Visual-inertial Navigation Method Of Four Rotor Aircaraft

visual and inertial combination as a navigation technology,is widely used in the mini-UAV autonomous flight control.The Kalman filter used can efficiently inference the latest state of Visual-Inertial navigation system,however,there is a large navigation bias during high-speed or long-term navigation.In this paper,batch optimization method is used to estimate the entire history state sequence of Visual-Inertial integrated system,and relinearize integrated system to overcome the problems of historical navigation error and linearized error accumulation of Visual-Inertial system,improve the navigation accuracy of quad-rotor Visual-Inertial navigation systemFirstly,the inertial and visual sampling data are aligned by pre-integrating the inertial measurement data with the camera sampling time scale as reference.In the process of optimization,the linear model of integral function is used to update the state of the system increment to avoid the integral operation.The principle of visual feature point matching tracking and direct method tracking feature point is analyzed and the direct visual odometer model is given.The direct method is used to track the ORB feature points to improve the stability in visual tracking.Secondly,based on the visual inertial observation model,the maximum a posteriori probabilistic model of Visual-Inertial navigation system is designed,and the incremental smoothing and mapping is used to estimate the state of Visual-Inertial navigation system.Finally,a four-rotor Visual-Inertial navigation system is set up to realize the Visual-Inertial navigation method.Three sets of Euroc datasets are used to test the Visual-Inertial navigation system based on the proposed method and the Kalman method.Experiments show that the Visual-Inertial navigation method in this paper can track the quadrotor trajectory better.Compared with the msckf method,the Visual-Inertial navigation method in this paper has higher navigation precision during fast flight and smaller accumulated error in the long-term navigation process.