PrivacyLens Delivers Truly Private Video Monitoring — By Replacing You with a Stick Figure
Sensing people with a thermal camera and replacing them in images on-device delivers actionable video footage without the privacy problems.
Researchers at the University of Michigan have come up with a new approach to in-home camera monitoring that serves to protect people's privacy — by turning them into anonymous stick figures.
"Most consumers do not think about what happens to the data collected by their favorite smart home devices," claims corresponding Alanson Sample of the problem his team set out to solve. "In most cases, raw audio, images and videos are being streamed off these devices to the manufacturers' cloud-based servers, regardless of whether or not the data is actually needed for the end application. A smart device that removes personally identifiable information before sensitive data is sent to private servers will be a far safer product than what we currently have."
That smart device is "PrivacyLens," which combines visible-light imagery with thermal imagery to figure out which parts of the image are people and which parts are background. Once a person has been spotted, the system blanks them out — replacing them with a stick-figure representation suitable for analysis for everything from activity recognition to fall detection, yet preserving their privacy.
"Cameras provide rich information to monitor health. It could help track exercise habits and other activities of daily living, or call for help when an elderly person falls," explains first author Yasha Iravantchi. "But this presents an ethical dilemma for people who would benefit from this technology. Without privacy mitigations, we present a situation where they must weigh giving up their privacy in exchange for good chronic care. This device could allow us to get valuable medical data while preserving patient privacy."
To further boost the user's control, PrivacyLens is adjustable: the system can be configured through a simple slider, blocking out people's faces in more public areas of a building but removing the whole person in more private areas. "There's a wide range of tasks where we want to know when people are present and what they are doing, but capturing their identity isn't helpful in performing the task," Iravantchi argues. "So why risk it?"
Iravantchi is set to present the work at the Privacy Enhancing Technologies Symposium in England this week; more information, including a PDF of the paper, is available in the event's proceedings.