Quick Initialization Method of Monocular VIO on MAV

The visual inertial odometry can effectively realize the autonomous pose estimation of the MAV, and the key to the robust operation and recovery of the VIO is the initialization process. In order to quickly and accurately provide the necessary initial value for the nonlinear optimization of the visual inertial odometry, this paper proposes a method that makes full use of the on-board sensor of the MAV to initialize the visual inertial odometry. This method uses a two-stage processor. In the MAV flight control system, the IMU deviation and initial attitude estimation are carried out through complementary filtering, and the basic principles of multi-view geometry are comprehensively used in the mission control computer to achieve pure visual SLAM. On this basis, the least squares method is used to realize the alignment of the world coordinate system and the camera coordinate system, and the external rotation parameters of the camera and the IMU are estimated online. The altitude change rate of the on-board altitude sensor of the UAV is used to initialize the camera scale information, so as to provide a comprehensive initial value for the nonlinear solution of the back-end of the visual inertial odometry. Through experimental verification, this method can meet the requirements of the initialization accuracy of the visual inertial odometer, accurately calculate the external parameters of rotation, stably and continuously estimate the posture of the MAV, and then realize the posture control of the MAV. The initialization method proposed in this paper takes about 6.7% less time than the classic method.