Monocular Vision Based Position Estimation And Navigation Of Quadrotor In Limited Space

Compared to fixed-wing aircraft,quadrotor have the unparalleled mobility and flexibility are widely used in both military and civilian.In this area,the more mature application is using GPS for precise localization and navigation,however,its easy to lose and instability of the GPS signals in the limited space,such as indoor and some intensive buildings.In this thesis,other stable and feasible methods are sought in order to maintain a stable flight position of quadrotor and realize its autonomous navigation.Vision sensors are favored because of its low power consumption and the ability to obtain the relative wealth of information.In this thesis,some research on a method of the monocular vision based position estimation and navigation which is suitable to the limited space are made.Besides,the research on algorithms,the designation and the development of software and hardware platform are completed.At last,it proposed a method to recover the failure of pose estimation which is based on Monte Carlo localization.The main work as follows:1.The application background and current research status in the vision-based pose estimation and navigation of quadrotor are explored in the thesis.According to the object of study,The feasible AR.Drone aircraft is selected as hardware platform and robots operation system are selected as software platform,then A dynamic model of quadrotor and the "black box" model based on system recognition are set up.2.The research on monocular vision based SLAM method.It’s including the calibration method of camera,feature detection method,matching feature points and tracking method,using epipolar geometry solve essential matrix,and the research on the method that split position tracking and mapping into two parallel thread.3.The scale estimate algorithm is estimated according to integration ultrasonic sensor with visual based method,and an alternative algorithm in large-scale spatial environment is proposed.The Kalman filter method is applied by combining the flight pose which is get from the vision SLAM method and the measurement of IMU to get a more accurate result.The aircraft can be controlled stably by designing four independently PID controller which is according to the four input parameters of the aircraft.4.Contrapose the problem that SLAM method relies on the surrounding environment which in some cases leads failures of the position tracking without any good recovery method.The method of position recovery based on Monte Carlo localization is proposed and validity of the algorithms proposed is verified through the computer simulation.