Marker-based tracking with unmanned aerial vehicles

With the availability of low-cost micro aerial vehicles (MAVs), unmanned aerial vehicles (UAVs) quickly gain popularity and application potential. This requires techniques that can be understood by non-experts and flexibly applied for rapid prototyping. Visual tracking is an essential task with many applications, such as autonomous navigation and scene acquisition. While marker-less methods emerge, marker-based methods still have major advantages including simplicity, robustness, accuracy and performance. In practice, however, multi-marker setups introduce complexity and calibration efforts that can void the advantages. This work proposes a solution for practical, robust and easy-to-use marker-based tracking with independent compound targets. We introduce two novel target designs and describe pose estimation, noise removal and geometric transformations. The concepts are implemented in a tracking library for the Parrot AR. Drone 2.0. We explain its video access and camera calibration, and provide a first set of intrinsic parameters, jointly estimated from 14 units with high accuracy and low variance. The library is applied in a one-day contest on automatic visual navigation of UAVs, where students without technical background and programming skills achieved learning by experience and rapid development. This shows the effectiveness of combining capability with simplicity, and provides a case study on robotics in interdisciplinary education.