In this paper we present a scheme for pedestrian tracking from an unmanned aerial vehicle (UAV), which includes the motion control of the UAV, and the visual tracking of a specific pedestrian from the moving platform. In the visual tracking part, we use an online updating feature queue and the Locality-constrained Linear Coding (LLC) method to match the pedestrian target. The ground station receives video stream from the UAV, and sends back commands to control the UAV's motion. If one pedestrian is specified as the target when the UAV is hovering, the UAV will track the pedestrian and keep a fixed distance from it. We only use visual information, which comes from the single front camera of the flying robot; no any GPS device is used. A Parrot AR Drone 2.0 and a Parrot Bebop 2 are adopted in the experiments, which are carried on in outdoor conditions, and experimental results verify the effectiveness of our scheme.