24 START-UP ALLEY Seervision
SEERVISION
COUNTRY: SWITZERLAND STARTED: 2016
Seervision’s autonomous robotic camera system won a Product of the Year award in the Best New AI/Machine Learning Technology category at this year’s NAB show in Las Vegas, but the start-up’s chief executive is at pains to stress that its success does not spell the end for camera operators. Nikos Kariotoglou, who co-founded the fledgling robotics outfit that meshes ML, computer vision and cinematography, says the tools his company offers merely enhance the camera operator’s role, enabling the production of higher quality footage. “We’re making a tool for the camera operator so they go from sitting behind one camera to sitting behind a fleet of cameras. The cameras are now intelligent and can do simple tasks themselves,” he explains. “On top of that, events with a strict budget will be able to producer higher quality footage – we’re calling it the democratisation of video production.” The company estimates that a three- camera set-up on a video production can save 30 to 40% of its budget using the Seervision system, which started life in the Automatic Control Laboratory of ETH Zurich. Kariotoglou, a former student there, enthusiastically describes it as, “the MIT of Europe”. The team started experimenting with filming lectures at the university, but soon moved onto more challenging scenarios, involving faster moving objects and multiple cameras. The company officially formed three years ago, funded by an ETH Zurich R&D grant and help from a handful of private financiers and accelerators. It has amassed an 18-strong team of computer vision, robotics and ML specialists.
RISE OF THE MACHINES Seervision promises to change things for productions by enabling AI-driven camera control
A partnership with a Zurich- based production company
also supported the firm’s growth by enabling the technologists to explore practical solutions the system could offer for live productions and events. The technology itself
“Perception involves understanding the scene around you, what the elevation
connects to the camera via its live SDI feed, allowing the operator to tap on the screen to define the type of shot – zoom, mid shot, close up, and so forth. The robot then takes these shots, saves them and tracks the objects as they move. From instruction to execution this action currently takes around three seconds, according to Kariotoglou. He adds that by training the robots to follow simple rules by learning from past footage, or by giving different cameras different tasks to perform, this part of the process will soon become completely automated. “Perfecting this part will lead to scalability, which will allow one person to handle 50 cameras – at the moment the limit is five or six,” he says. Using data to train models to recognise production rules through the combination of cinematography and machine learning forms one of the three key pillars of Seervision’s technology. The other two, according to Kariotoglou, are a focus on perception and on control, and are based on similar principals to those used by autonomous car systems.
of the object is in respect to the camera. It also identifies where you are in respect to the environment and the object you are filming,” he explains. “Then you have control – how the camera moves in a way that feels human. If you move a camera too fast, you feel it in the images. Like if a car suddenly accelerates abruptly, you feel it as a passenger. So you are trying to make the operation of the camera indistinguishable from human operation control.” The company is adopting an automation-as-service model, charging a licence fee for the usage of the system, with the client able to purchase additional or bespoke features for their production as and when they require. There is also a one-off payment for installing the processing hardware and the robot, if the client doesn’t already own one. The camera systems are currently being deployed to add value to corporate events. Sport is also a key area, and especially relevant given that Seervision is based in a nation that’s home to around 30 international sports organisations. “We’re in talks to capture sports events, in football and tennis, for the lower leagues, in higher quality. To be honest we are not there yet with the technology, but we’re partnering up with everyone who approaches us and are working on a prototype,” says Kariotoglou. He adds that Seervision hopes to have a working model for sport ready by the end of 2020.
YOU ARE TRYING TO MAKE THE OPERATION OF THE CAMERA INDISTINGUISHABLE FROMHUMAN OPERATION CONTROL
feedzinesocial feedzinesocial feedmagazine.tv
Powered by FlippingBook