In recent years, quadrocopters – which are small helicopters with four rotors – have experienced a surge in popularity. They are no longer only used as a research platform or high-tech toys, but also commercially for real-world applications, such as for aerial video recording with spectacular camera stunts, for inspection tasks, or for the surveillance of hazardous situations.
The Computer Vision Group at TUM investigates methods which allow a quadrocopter to fly autonomously – that is without human intervention – in indoor and outdoor environments. The goal is for the quadrocopter to act completely independently, without requiring external sensors such as GPS or a prior map of the environment.
All information necessary for flying is acquired by the quadrocopter and its on-board sensors. This is achieved by a multitude of different sensor modalities. Cameras localize the quadrocopter and allow the creation of highly accurate 3D reconstructions of the environment, while accelerometers and gyroscopes provide the necessary stability.
One of the main challenges when working with flying robots lies in the limited payload: In this respect, cameras are the ideal sensors, as they provide rich amounts of information while being extremely lightweight and small, and have low power consumption. In order to process this information, modern, efficient and innovative methods from the field of computer vision are required. Our group actively works on topics such asdense visualodometry,SLAM, and image-based 3D-reconstruction.At first, such methods can be used to assist a human pilot, for example, by automatically detecting and avoiding collisions with obstacles, or by compensating for disturbances such as wind.
In the future, a large variety of different tasks could be fulfilled by quadrocopters: Architects, for example, could quickly and easily acquire 3D models of buildings. After a storm, quadrocopters could automatically scan roofs in residential areas for damages. But also for private users flying cameras are utterly interesting: Imagine you could film yourself during your next ski tour next winter.
The system shown in this video was recently developed at the Computer Vision Group at TUM: It allows a quadrocopter to estimate and hold its position or to fly simple figures using a single, forward-directed on-board camera. The results of this project will be published at the IEEE International Conference on Intelligent Robots and Systems (IROS) which takes place this year from October 7-12 in Portugal. Furthermore, a live demonstration of the system will be given at the TUM Open Day on October 27, 2012 in the main hall of the computer science building. Those who are interested to learn more about this topic are cordially invited to participate in the lecture “Visual Navigation for Flying Robots” of Dr. Jürgen Sturm, which takes place for the next time in summer term 2013.
Do you want to know more? Visit our research page.