Target Tracking and Following from a Multirotor UAV

In this paper, we present a general target tracking and following architecture for multirotor unmanned aerial vehicles (UAVs), and provide pointers to related work. We restrict our discussions to tracking ground-based objects using a monocular camera that is not gimballed. Target tracking is accomplished using a novel visual front end combined with a recently developed multiple target tracking back end. The target following is accomplished using a novel, nonlinear control algorithm. We present an end-to-end target tracking and following architecture that uses a visual front end to obtain measurements and compute the homography, a tracking algorithm called Recursive Random Sample Consensus (R-RANSAC) to perform track initialization and tracking, a track selection scheme, and a target-following controller. Target tracking and following is accomplished using a monocular camera, an inertial measurement unit, an on-board computer, a flight control unit, a sensor to measure altitude, and a multirotor UAV under the assumption that the target is moving on fairly planar ground with nearly constant velocity.