Pose Estimation Of Multi-sensor Fusion For Quadrotor Systems

The development of quadrotors has been extremely rapid in recent years.In order to achieve industrial applications,the pose estimation of quadrotor has become one of the current research hotspots.This thesis focuses on odometer methods that fuse visual and inertial information.We aim to.achieve its real-time computing via onboard processor with strong robustness and high accuracy.First,one of the research scenarios in this thesis is the Mission 7 of International Aerial Robotics Competition(IARC),in which the quadrotors are required to use their own sensors in a GPS-denied environment to achieve pose estimation of themselves and the target.Through the color features,the target and background are separated,and the physical size of the target is well preserved.Based on the prior knowledge of the known target actual size and the perspective-n-point(PNP)method,the pose of the quadrotor in the target frame is obtained by using the Levenberg-Marquardt(LM)algorithm.Next,the IMU information is fused to solve the pose of the quadrotor and the target in the arena frame.Finally,the quadrotor realizes the target detection and tracking by using the pose information.Then,in order to realize pose estimation of quadrotor under normal circumstances,a pose estimation algorithm based on ARTag is proposed.The algorithm can not only robustly identify a single marker,but also jointly solve all the feature points of the markers observed in the same frame.The flight can still be stable in the case of 0.5m/s.It is an improved vision odometry technique based on the feature point.The accuracy of real-time pose estimation can fully meet the requirements of flight.Due to the known size of the ARTag,monocular vision can be used,which reduces the computational and loading burden on the quadrotor.Finally,aiming to solve the error caused by image distortion and the observation of less markers,a pose estimation algorithm based on multi-sensor fusion is proposed.The author fuses the IMU data of quadrotor and the pose estimation method based on ARTag,and designs a modular Kalman filter framework.In the filtering framework,the visual part is no longer a black box,but the observed feature points are added to the state vector for state iteration.A new time synchronization algorithm is designed to match the data with different frame rates reasonably.