Title: Real-Time Implementation of Vision-Aided Navigation for Small Fixed-Wing Unmanned Aerial Systems
Author(s): Timothy Machin, John Raquet, David Jacques, Donald Venable
Published in: Proceedings of the 29th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2016)
September 12 - 16, 2016
Oregon Convention Center
Portland, Oregon
Pages: 1305 - 1311
Cite this article: Machin, Timothy, Raquet, John, Jacques, David, Venable, Donald, "Real-Time Implementation of Vision-Aided Navigation for Small Fixed-Wing Unmanned Aerial Systems," Proceedings of the 29th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2016), Portland, Oregon, September 2016, pp. 1305-1311.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In
Abstract: The goal of this project was to develop and implement algorithms to demonstrate real-time positioning of a UAV using a monocular camera combined with previously collected orthorectified imagery. Unlike previous tests, this project did not utilize a full inertial navigation system (INS) for attitude, and instead relied on the attitude obtained by an inexpensive commercial-off-the-shelf (COTS) autopilot system. The system consisted of a machine vision camera, a Pixhawk autopilot, and an Intel NUC. The components were installed in a 12-ft Telemaster aircraft and flown over Camp Atterbury, IN over a sequence of flight tests. These flight tests were used to verify the real-time performance of the vision-aided navigation solution. The vision algorithm used SIFT to identify features in the image coming from the onboard camera, and matched those with the feature database using the Fast Library for Approximate Nearest-Neighbors (FLANN). Once features were matched, the algorithm attempted to solve a six degree-of freedom (6DOF) solution (Full position and attitude), and a three degree-of-freedom (3DOF) solution (Position only, with attitude obtained from the Pixhawk autopilot). These solutions used the Perspective-n-Point (PnP) algorithm to compute the absolute position of the vehicle. The system obtained valid solutions over much of the flight path, which consisted of orbits in and around the Himsel airfield at Camp Atterbury. Prior to the test, it was uncertain whether the relatively poor attitude accuracy of the Pixhawk autopilot would be of sufficient accuracy to aid the solution. The tests demonstrated that such aiding is beneficial, reducing the position accuracy (horizontal DRMS) of the 6DOF solution of 54.8 meters, to the 3DOF solution (with attitude aiding) DRMS of 14.5 meters in real-time. Some calibration errors were found in the real-time results, and after correcting for these errors in post-processing, a final DRMS accuracy of 8.2 meters was obtained. Post-processing analysis was also performed to test variants to the feature database used.