Flexible vision-based navigation system for unmanned aerial vehicles Erik Blasch Georgia Institute of Technology George W. Woodruff School of Mechanical Engineering Atlanta, Georgia 30332 ABSTRACT A critical component of unmanned aerial vehicles is the navigation system which provides position and velocity feedback for autonomous control. The Georgia Tech Aerial Robotics navigational system (NavSys) consists of four DVTStinger7OC Integrated Vision Units (IVUs) with CCD-head panning platforms, software, and a fiducial onboard the vehicle. The IVUs independently scan for the retro-reflective bar-code fiducial while the NavSys image processing software performs a gradient threshold followed by a image search localization of three vertical bar-code lines. Using the (x,y) image coordinate and CCD angle, the NavSys triangulates the fiducial's (x,y) position, differentiates for velocity, and relays the information to the helicopter controller, which independently determines the z direction with an onboard altimeter. System flexibility is demonstrated by recognition of different fiducial shapes, night and day time operation, and is being extended to on-board and off-board navigation of aerial and ground vehicles. The navigation design provides a real-time, inexpensive, and effective system for determining the (x,y) position of the aerial vehicle with updates generated every 51 ms (19.6 Hz) at an accuracy ofapproximately in. Keywords: machine vision, navigation, unmanned aerial vehicles, helicopter control 1. INTRODUCTION Position measurement is essential in guiding autonomous vehicles as demonstrated by research on an unmanned aerial vehicle (UAV). Machine vision, where images are digitized and analyzed by micro-computers, is a promising technology which provides accurate and reliable estimates of position for feedback control of vehicle movement. The machine vision system (MVS) may be located either on the vehicle or strategically located close to the vehicle's operating environment. MVS outputs can be used to estimate vehicle position and velocity, compensate for disturbance inputs, identify and avoid obstacles, and determine task completion status. Since, a MVS has the capability to sense a minute change with in its field of view, a vision system and UAV are excellent for high-risk security surveillance such as aerial observation for the military. During peace-time or war-time operations, the personnel cost to the military is high, and losing a soldier for information gathering far outweighs a lost vehicle. Another use for UAVs is the observation, equipment retrieval, and deposit of equipment/resources in hazardous situations. A chemical rail turnover and a contaminated nuclear facility provide two such examples where human exposure is considered unacceptable. As a demonstration of potential UAV uses, the Association for Unmanned Vehicle Systems holds a contest which challenges collegiate students to design an autonomous vehicle to pick up and relocate objects. This paper details a low-cost machine vision system for use in autonomous aerial vehicle control by presenting its use for the Georgia Tech Aerial Robotics Team System (GTARS), equipment characteristics, fiducial-tracking algorithms, and results of position measurement. 2. GEORGIA TECH AERIAL ROBOTICS The GTARS team is conducting state of the art research on autonomous control of UAVs. The GTARS' vehicle is a 58 /SPIE Vol. 2352 Mobile Robots IX (1994) O-8194-1687-8/95/$6.OO Downloaded From: http://spiedigitallibrary.org/ on 09/25/2013 Terms of Use: http://spiedl.org/terms