Fast Object Tracking Using Micro Aerial Vehicles

摘要

The development of robust and applicable visual object tracking algorithms for surveillance purposes is among the most studied applications of aerial robots. The tracking problem is mostly considered as an open loop system, with no control on the tracker camera motion. In the few exiting closed-loop approaches, the visual feedback is typically expressed in pixel coordinates of image, leading to poor performance in fast motions due to the coupling problems of flying robot’s states. In this paper, we introduce a fast motion planning for the flying camera that is based on visual tracking feedback in the inertial coordinates by incorporating kinematics model of the robot in motion planning. The proposed strategy is shown to resolve visual object tracking challenges by maintaining object scale and perspective, and reducing the chance of occlusion and illumination change, thus increasing the tracking durability. As a result, the flying robot is capable of tracking a person running at speeds up to 5 meters per second, with no predetermined paths, a task which requires rather visually challenging maneuvers for the flying robot.