Engineering students from Canada’s Simon Fraser University (SFU) have developed an artificial intelligence (AI)-driven privacy shield that scrambles the images of occupants captured by smart home security systems vulnerable to hacking.

Smart home security systems like Amazon’s Ring and Google’s Nest are growing in popularity, and corresponding reports state that such systems have been intruded by hackers, allowing them to view images of unaware occupants in their homes. SFU students Saeed Ranjbar Alvar and Hyomin Choi developed a computer-vision algorithm that scrambles images as video is uploaded to storage platforms, similar to those used in smart-home security systems.

Alvar and Choi altered a video compression algorithm to blur out content, making it appear scrambled to the human eye. They then incorporated machine learning techniques to train the AI model to identify segments of the video’s data stream, or bitstream, that correspond to human faces. Although the video looks scrambled to humans, the AI system can locate human faces within the video and follow their movements throughout the video. Authorized moderators can still detect the presence of human subjects onscreen with assistance from the AI model, without actually seeing them, making the data still useful for law enforcement and security providers despite being blurred. In other words, security providers can distinguish occupants from intruders by applying the software to reveal the identity of those in the home, while hackers and voyeurs only see scrambled images. Currently, the model identifies faces in the same way that standard facial recognition scanners do.

“When you upload a video or your smart-home system streams to your security provider, the image itself is apparent and the server that holds the image’s data can identify you,” explained Alvar, an SFU engineering science Ph.D. student. “We see this when we upload to Google or Facebook and it suggests who’s in a photo using automatic facial recognition software. Smart-home security systems do this too, to protect us from intruders.”

The team sees a host of potential applications for the privacy software including in taxi cabs, assisted living facilities and other locations where video still needs to offer enough data for analysis, but without jeopardizing the privacy of those appearing on camera.