Pose Estimation And Control Of Quadrotors Based On Monocular Vision

Research interest in autonomous micro-aerial vehicles has grown rapidly in the past years. Accurate positioning is the basis to realize complex tasks. Currently, positioning of quadrotors can be categorized into two main research approaches. One relys on external systems to provide accurate positioning information, such as GPS and motion capture system. However, motion capture system can only be used in the lab environment and GPS almost loses efficacy in the places such as dense buildings, indoors and tunnels. The other is based on simultaneous localization and mapping(SLAM) method. In this dissertation, we mainly studied the positioning of quadrotors based on monocular SLAM in the unknown environment and flight control based on visual positioning information.The system of quadrotors positioning based on monocular vision was established in this dissertation. The hardware part was composed of a quadrotors and a laptop computer. And the software part consisted of three components: a monocular SLAM system, an extended Kalman filter, and a PID controller. The monocular SLAM was used to estimate the visual pose of quadrotors. EKF is employed to provide data fusion and compensate for time delays in the system. We applied PID control to steer the quadrotors to reach the desired goal location. Positioning system was carried on the embedded transplantation and implemented airborne online processing. At the same time, the airborne system also solved the positioning inaccurate problem affected by the wireless signal. Self-similarity features in high spatial frequency were omnipresent. For high frequency self-similar feature points easily caused false matches, we optimized the pose algorithm to solve this problem by only utilizing the features extracted in the higher pyramid levels during the map-building process, while utilizing the lowest pyramid level features in the tracking process.In order to verify the validity of the method, comparative experiments of different methods were carried on in this dissertation. The experiments demonstrated that the pose estimation and control system of the quadrotors performed well in desired path tracking. Through using airborne processors, the positioning system was small in size and portable. It could be widely used in many fields.