First person video appears to be a perfect way to broadcast footage without revealing an individual’s identity. It’s intimate but anonymous: hiding the filmmaker’s face and body while showing the world through their eyes. But now, as GoPros, Google Glass, and police body cameras make this sort of footage more common, Professor Shmuel Peleg and Yedid Hoshen of the Hebrew University of Jerusalem have created a method of identifying first-person filmmakers solely from the wobble of their cameras.
"The idea of distinguishing one person from another comes from analyzing the person’s gait," explains Professor Peleg in an interview with The Verge. "Everyone has a different body build, height, muscles, and skeleton, and so the body moves differently and uniquely."
Peleg and Hoshen found that these biometric markers convert neatly into distinctive camera movements, publishing their results in a paper titled Egocentric Video Biometrics. By tracking the "optical flow" of objects and surfaces from one frame to the next they were able to determine volunteers’ identity from just four seconds of first-person footage.
"The fact that their face isn’t seen doesn’t mean that they are anonymous."
"This is like a fingerprint," says Peleg. "In order to find the person you have to have their fingerprint beforehand. But we can compare two people and say whether two videos were shot by the same person or not." He says that his main motivation in publishing the findings was to "make people think twice before uploading egocentric footage to the web, adding: "The fact that their face isn’t seen doesn’t mean that they are anonymous."
Peleg and Hoshen explain that in the future, law enforcement agencies might even be able to link first-person footage to video captured by CCTV. "Though we haven’t done this form of recognition, when you look at a person from a surveillance camera you can see the way they move and the way they move their head," says Peleg, adding it could be "possible" to connect the two, especially in surveillance-heavy countries like the UK where there is one CCTV camera for every 11 people.
Although their research only looked at head-mounted cameras, Peleg and Hoshen are confident the same methods could be used for body-mounted devices: an important development considering President Obama’s recent pledge to purchase 50,000 body cameras for US law enforcement. In the future, confirming that certain footage was filmed by one individual and not another could be as important for holding law enforcement accountable as it is for bringing criminals to justice.