UAV for indoor navigation using a single camera

• Main Author: Lau, Sin Ye
• Other Authors: Wang Han
• Format: Final Year Project
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/65787
• Institution: Nanyang Technological University

This report presents a system that enables autonomous localization and navigation of a quadrocopter based on visual feedback from the monocular camera. The system uses the quadrocopter on-board sensors and artificial makers or tags. Since the drone does not possess the human-intelligence to differentiate the different types of environment it is in, one of the solutions is to use artificial markers. The drone will read the makers with the vertical camera to localize itself and navigate along a flight path in an enclosed environment. In constraint spaces there will be areas of blind spots along the flight path and possible communication delays. Therefore to tackle this problem, an Extended Kalman Filter (EKF) algorithm was implemented in the system to combine the data received from other on-board sensors such as the visual odometer and gyroscopes to compensate for the temporary loss of visual tracking and communication delays. A proportional-integral-derivate controller (or PID controller) was implemented in the system to control the drone’s position and orientation. A graphical user interface (GUI) was designed to be user friendly to display real time data received from the drone. To control an unmanned aerial vehicle (UAV) in semi-autonomous mode, a user has two available options either by the keyboard or play station remote control. To determine the responsiveness of an AR Drone manufactured by Parrot in the presence of external disturbances, a number of autonomous flight programs were designed to carry out different tests in-order to understand more and make further improvements. In different occasions, the AR Drone demonstrates it is able to rely on the inertial measurements despite the temporary loss of visual tracking and communication delays. However, the predictive model and compensating of the communication delays will cause significant effect in the accuracy of the state estimation.