Vision-Based Object Path Following on a Quadcopter for GPS-Denied Environments

This paper presents the development of an object path following algorithm for a quadcopter using on-board vision and localization. This development aims to increase capabilities of autonomous unmanned aerial vehicles (UAV), with the use of object path following. By following an object's path, reliance on obstacle avoidance strategies can be reduced, under the assumption that the object's path is free of static obstacles. Relative position of the target has been estimated using on-board processing and monocular vision. On-board localization has been achieved through dead-reckoning, where velocity estimates are calculated from optical flow. Horizontal positional estimates with a standard deviation of 6.5cm have been achieved experimentally with this method. The use of optical flow allows for localization in GPS-denied environments. Both the position and path following can successfully run in moderate winds and variable lighting. The UAV can station keep to within 40 cm of desired position. The object's path can be followed with an rms error of 28 cm and maximum deviation of 78 cm.