Computer system turns drones into cinematographers
When drones are used for aerial shots in movies, there's typically one person remotely operating the aircraft, while another remotely controls its camera. Thanks to new research from MIT and ETH Zurich, however, it may soon be possible for drones to autonomously follow along with an actor, keeping their face framed "just right" the whole time – while also avoiding hitting any obstacles.
To utilize the experimental new system, operators start by using a computer interface to indicate who the drone should be tracking, how much of the screen their face or body should occupy, where in the screen they should be, and how they should be oriented toward the camera (choices include straight on, profile, three-quarter view from either side, or over the shoulder).
Once the drone is set in action, the computer wirelessly sends it control signals that allow it to fly along with the actor as they walk, adjusting its flight in order to maintain the shot parameters. This means that if the actor were to start turning their back on the drone, for instance, it would automatically fly around in front of them, to keep their face in the shot. Likewise, if they started walking faster, the drone would also speed up in order to keep them the same distance from the camera.
It's additionally possible for the aircraft to follow small groups of actors, working to keep that group framed a certain way within the shot. The user can stipulate one of those actors as the main subject, ensuring that the drone moves in order to keep other actors from blocking the camera's view of them.
The system utilizes algorithms that predict the actor's trajectory about 50 times a second, allowing the aircraft to effectively stay one step ahead of the action. This also allows it to correct its own flight path if its onboard sensors detect that it's heading toward a stationary obstacle, or if a moving obstacle (such as an actor) is on a collision course with it.
A team led by MIT's Prof. Daniela Rus will be presenting a paper on the research later this month at the International Conference on Robotics and Automation. The system is demonstrated in the video below.