RGB-D SLAM based on quadrotor platform

This thesis takes the UAV-based RGB-D SLAM system as the object to deeply study and improve the existing visual-SLAM method VINS, and construct a real-time RGB-D SLAM system based on quadrotor flight. It demonstrates an efficient and robust RGB-D SLAM system, which fuses RGB-D camera and IMU measurements, and uses a tightly coupled nonlinear optimization to obtain accurate pose estimates. First, the thesis summarizes the research status, analyzes the architecture of the keyframe-based visual inertial navigation system in modules, and introduces the principle of the measurement preprocessing, system initialization, and visual inertial state estimator. For the difficulty in initialization, we uses the depth camera measurement to add a robust and stable static initialization, which ensures that the UAV can obtain a stable state estimation before takeoff. And we also solve the problem of feature gathering in the frontend, making feature point distribution uniform, and propose a judgment to determine weak texture/blur regions to increase feature point stability. In addition, by changing the feature point type, we improve the running speed of the frontend to ensure real-time performance. Then, the thesis systematically analyzes the kinematics and control theory of UAV, describes the basic principles of UAV trajectory planning, and constructs the minimum jerk problem to generate a smooth trajectory suitable for UAV motion. Meanwhile, it designs a flight controller for geometric control in SE(3) space, and the convergence of partial control theory is proved by Lyapunov equation. Finally, in order to verify the above improvements, the thesis conducts experiments of feature extraction improvement, feature tracking improvement, static initialization and indoor real-time localization, respectively. And the results successfully verify the effectiveness and superiority of the proposed system. The real-time RGB-D SLAM system proposed in this thesis is expected to provide an effective and robust online state estimation for UAV as feedback information for the control system to implement autonomous flight in unfamiliar environments. Keywords: RGB-D SLAM, UAV, quadrotor, VIO, VINS, state estimatorMaster of Science (Computer Control and Automation