Videoconferencing privacy app lets you decide what others see of your surroundings
Devices equipped with cameras have given us the power to instantly share ideas and images, but have also made it very easy to leak sensitive information. In order to prevent future disasters, researchers at Duke University have designed an app that lets users specify what others can and cannot see during a video conference. Users simply draw a rectangular border around the scene deemed safe to share with others and the surrounding areas are completely blocked out from view.
'There are more and more cameras every year. They're incredibly useful,' said Landon Cox, an associate professor of computer science at Duke University. 'But the downside is we're now converting large swaths of our surroundings to a digital format that's easy to access and share, including things we might not want to be digitizing.' The simplest way to ensure privacy, is to disable the camera or microphone when sensitive information is in the frame, Cox said.
Prior to Duke's work, other researchers have developed technology that predicts what information a user will want to hide and add software blurs in each frame – this is known as 'blacklist'. However, developers have to create an extensive list of objects that could be potential threats if seen by others. 'Things that some people consider sensitive might not be sensitive to you,' said assistant professor of computer science Ashwin Machanavajjhala, who co-authored the research.
'It's hard to build something that covers all possible scenarios.' Although blacklist has created a long list of objects deemed troublesome, the researchers at Duke understand the task of building software that can quickly and consistently conceal and detected them can be challenging. And even if it is fails just one or 2 percent of the time, the team knows it is not secure enough for the masses.
Instead of anticipating and predicting what objects can or cannot be seen, the researchers are letting the user decide – and has create two examples to demonstrate their technology. One is designed to protect sensitive information on two-dimensional surfaces such as whiteboards and computer presentation slides.
The other safeguards images of three-dimensional objects such as keyboards and faces. However, both examples work the same – users select the part of a screen that is 'OK' to share by drawing a rectangular border around it, either by hand or with a few clicks of a mouse.
And once the software understands what it is looking at, it intercepts all incoming frames from the video stream and rapidly scans frame by frame for a match using computer vision technology. Only authorized objects are allowed to pass from the camera to third-party software, like smartphone apps --everything else is blocked out by default. 'The key challenges in designing these systems were to ensure that the marking process was easy for the users, and that detecting public regions did not slow down the camera output or the smartphone,' Machanavajjhala said.
To test their software, the team asked 26 people to use an Android smartphone to scan QR codes – both with and without the security features. The participants then reported back with how they felt about the speed and ease of the cameras on the 'secure' smartphones compared to the unmodified smartphones. The team also tested their security software on videos that were shot when the camera was in motion, which they found it could reliably safeguard sensitive regions while still delivering 24 frames per second.
This is considered fast enough for human eyes to perceive a smooth moving picture rather than a flickering image on a screen. This isn't the ultimate solution to image privacy protection, Machanavajjhala said.
The software doesn't protect things caught by cameras outside a person's control, for example. 'If you're just walking around on the street, and you want to ensure that your face isn't captured, this won't work,' he said. But the main goal looking to the future is to encourage technologies to design and develop products that give users more privacy when it comes to their own devices, the team said.
