Quadcopter Capable of Autonomously Chasing Micro-Aircraft with Real-Time Visual Tracker

Abstract

It is difficult for unmanned aerial vehicles to chase another micro-aircraft (MA) due to the small size and its fast maneuverability. Thus, this study developed a fast and accurate visual tracker for real-time inference. Then, a quadcopter was controlled to chase a target MA by considering the result of the visual tracker. Specifically, the pitch, throttle, and yaw of the quadcopter were determined by the PD controller based on the position, and the size of the MA in the image. The newly developed visual tracker comprises an adaptive search region (SR) and a fully convolutional neural network. The size and the location of the SR were constantly adjusted over image frames by considering the tracking result of the MA in previous frames. Furthermore, if the size and the location of the SR are not precise enough, the SR was updated to minimize the tracking failure. Performance of the SR was improved by using the Kalman filter. In real flight experiments, the quadcopter equipped with the proposed model successfully chased the MA which randomly moved at approximately 5 m/s.