Pose Estimation Based On The Fusion Of Visual And Inertial Information For MAVs

Accurate pose estimation is the key to realize autonomous flight control for MAVs.At present,most MAVs integrate the global satellite navigation system(GNSS)and inertial measurement unit(IMU)to realize the pose estimation,but it can not work normally under the GNSS denied condition.The pose estimation based on visual information can solve the localization problem of MAVs when the satellite signal is not available.The pose estimation algorithm based on monocular vision and IMU is researched,and a pose estimation system is realized to fuse the information from monocular vision and IMU.Firstly,a multi-sensor fusion pose estimation system is designed,which includes monocular camera,IMU and ultrasonic sensor.The problem of multi-sensor system initialization is solved,including IMU alignment aided by the magnetometer,IMU initial drift estimation by Kalman filtering algorithm,and the calibration of the intrinsic parameters of monocular vision and the extrinsic parameters between monocular vision and the IMU.Secondly,to deal with the problem that monocular vision can not obtain absolute scale in the process of pose estimation,the ultrasonic sensor and IMU are used to estimate the scale information of monocular ORB-SLAM algorithm in different environment.In the ORB-SLAM initialization process,the least squares algorithm is used to estimate the absolute scale from the acquired data from the ultrasonic sensor in the indoor environment;six degrees of freedom pose information calculated from the IMU is used to correct the scale of ORB-SLAM in the outdoor environment.Finally,the pose information obtained by the monocular ORB-SLAM algorithm and the IMU is fused by the Extended Kalman Filter(EKF),thus the output frequency and the accuracy of the whole pose estimation system are improved to make the system applicable on the small rotor MAVs with high maneuverability.The information fusion process is as follows: firstly,the system equation is established according to the kinematic model of the IMU,and the inertial navigation solution is acquired;and then the error state equation is derived to model the system error state;finally the pose estimation result obtained by the monocular ORB-SLAM is used as the measurement,and the optimal estimation of the system error is acquired by the EKF to correct the pose estimation result by the IMU.In order to verify the effectiveness of the proposed algorithm,the experiments were conducted in the indoor and outdoor environments.The results of several different experiments in different environments show that the proposed algorithm can accurately estimate the scale to obtain absolute scale information of the monocular SLAM,and accurate pose estimation results can be obtained using the proposed algorithm based on multi sensor information fusion,which meets the demands of MAVs in the speed and accuracy of pose estimation.