Vision-based Tracking by a Quadrotor on ROS * * This work was supported in part by the U.S. Army Research Office under grant W911NF1510275.

摘要

Real-time following of a moving target by a flying tracker without radar detection and GPS reception inspires vision-based tracking through the use of surveillance camera, laser-scanned lidar, and SLAM (simultaneous localization and mapping). In this paper, we propose vision-based tracking of a moving target in three consecutive steps, image processing, coordinate transformation and trajectory following. A testing platform containing two spawned quadrotors as target and tracker is developed on ROS, from which sequence of images are generated by an on-board camera of the tracker while the target moves along a rectangular path on the ground. An image processing algorithm based on target recognition is proposed to achieve visual tracking on image frames. Coordinate transformation is established to recover the original target trajectory as the desired path of the tracker in the real world from the image processing algorithm. The target following control law then moves the tracker along the desired trajectory point-by-point through velocity control.