Quadrotors are agile and flexible platforms for both academic research and industrial applications, and are capable of operating tasks in challenging environments for humans. Many of these tasks require quadrotors to track a moving target on the ground in complex environments where GNSS is unavailable. This thesis addresses the issue of autonomously tracking a moving target for a quadrotor in GNSS-denied environments. The main contributions of the thesis are summarized as follows.
1. This thesis implements an optimization-based visual-inertial odometry for state estimation of the quadrotor. Collecting measurements from the camera and the IMU, the camera residual is obtained by the reprojection error of the feature points, and the IMU residual by the preintegration error. The visual-inertial odometry estimates the states of the quadrotor by jointly minimizing the camera residual and the IMU residual and solving a nonlinear least square problem. To maintain a real-time performance of the visual-inertial odometry, the optimization is limited on a sliding window of a fixed size by marginalization.
2. This thesis presents a trajectory generation method based on piecewise Bézier curve, which plans safe, smooth and dynamically feasible motions for quadrotors. The trajectory of the quadrotor in the 3D space is represented as the piecewise Bézier curve using Bernstein polynomial basis, and generated by minimizing the control effort on the trajectory of the quadrotor as the cost function. With the special properties of Bézier curves, the position of the entire trajectory and its derivatives can be bounded within the safe and feasible spaces by the control points. Therefore, the motion constraints of the trajectory are enforced in the form of linear equalities and inequlities. Combining the cost function and the linear constraints of the optimal trajectory, the trajectory of the quadrotor is generated by solving a quadratic programming.
3. This thesis proposes a vision-based moving target tracking system of quadrotors with visual-inertial localization. The visual-inertial odometry is utilized to estimate the states of the UAV by fusing visual and inertial measurements, and the states of the target are estimated by extended Kalman Filter from visual detection. This research formulates the target tracking problem as optimization-based trajectory generation, and a weighted sum cost function jointly penalizes the tracking error, the control cost of the trajectory and the trajectory length, while the safety and feasibility constraints are enforced. Simulations and real-world experiments are conducted to validate the effectiveness of the system.
修改评论