Recent years have seen an explosion of what are known as “egocentric” videos — videos taken from cameras mounted to a person’s head or body. These videos have a striking appeal because of two contrasting qualities. They are both personal — viewers get to see the world through the filmmaker’s eyes — and eerily anonymous: With a head-mounted camera, you never see the face of the person shooting the footage.
That anonymity may be slipping away, however. Yedid Hoshen and Shmuel Peleg, computer vision researchers at Hebrew University in Israel, recently posted a paper describing a method for zeroing in on the identity of an egocentric filmmaker. It involves computer analysis of the person’s motion patterns, which reflect the distinctive biometric features of each filmmaker, like height, stride duration, and the flexibility of the neck.
“It’s possible every single person would have a different biometric signature that would be encoded into these egocentric videos,” says Hoshen.
He and Peleg had 34 participants take two videos each using GoPro cameras mounted to baseball caps. They trained a computer-based neural network to split frames from those videos into blocks, which became the unit of analysis. Using blocks from only four seconds of video footage, the neural network was able to determine with 90 percent accuracy which pairs of videos were created by the same filmmaker.
Given the range of things filmed “egocentrically” -- not just skydives and ski runs, but the raid that killed Osama bin Laden, and masked commandos in training -- there could be a lot at stake at breaking through their anonymity.
“Say in one of the videos [the filmer] doesn’t have the mask and you can see his face in the mirror, and in another video, he does have a mask,” says Hoshen. He says voice recognition software and data from cellphone records might be other ways to triangulate a person’s identity from a handful of linked, but anonymous videos.
The loss of anonymity is often regarded as a bad thing, but there may be advantages to this kind of identification scheme. Egocentric cameras are likely to proliferate as wearable technology becomes common, and the distinctive features of your gait could be a way to lock your device, so that only someone who walks just like you can use it. There may be applications in law enforcement, too: Biometric analysis could be used to verify that a specific egocentric video was in fact taken by a specific officer.
For the researchers, however, the next step in their research program is to better understand the power of this kind of analysis. They know it provides some ability to identify egocentric filmmakers, but with a much larger dataset from someplace like YouTube or Facebook, they could determine if the way we shoot GoPro footage is as distinctive as our fingerprints.
Kevin Hartnett is a writer in South Carolina. He can be reached at email@example.com.