Camera tech CamPro prevents unauthorized recognition

Facial recognition enables camera surveillance to identify people based on their faces. It has various purposes, such as finding missing people, identifying people at airports, and unlocking smartphones. However, it threatens personal privacy as it can track and profile people without their consent.

That is why Zhejiang University researchers created CamPro, a camera-sensor-based anti-facial recognition system that modifies images at the source. It causes cameras to obscure a person’s facial features while leaving enough data to support other non-sensitive vision applications such as person detection.

In other words, it hides your face while letting authorities identify you with other details.

How does this camera tech work?

Camera Privacy Protection modifies photos and videos at the source to follow a “privacy-preserving by birth” approach. It exploits the tunable settings of the image signal processor (ISP), a hardware part that turns raw camera data into a standard image format.

The ISP typically has adjustable parameters that affect an image’s color, brightness, sharpness, contrast, and noise. CamPro can produce unreadable images by altering these.

However, it still allows others to identify someone through non-visual means, such as person detection or pose estimation. As a result, it protects personal privacy while allowing authorities to spot people in case of emergencies. 

Researchers say they tested CamPro on widely available cameras like webcams, surveillance cams, and smartphones. Consequently, they proved it can prevent facial recognition from various cutting-edge models. 

READ: AI editor turns microbes into renewable fuels

Moreover, the Chinese researchers demonstrated that CamPro can maintain the functionality of other computer vision applications like object detection and activity recognition. 

CamPro doesn’t require additional software or hardware to integrate into existing camera modules. However, this camera tech has some drawbacks. 

For example, it may not function in cameras with fixed or non-tunable iSPs or with low-quality sensors. Also, CamPro may produce visual artifacts or distortions that may affect a picture’s aesthetics. Learn more about this research on its arXiv page. 

Other camera surveillance methods

Cameras don’t need to see you to track your activities. MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) researchers proved someone could use AI on camera sensors to image movements. 

I discussed this research in my other article. Most smartphones and tablets use these light detectors to trigger specific features, such as dimming your screen in the dark. 

Modern sensors are so sensitive that they can identify subtle changes in brightness. It’s like surrounding objects produce shadows that the light sensor can discern. 

Yang Liu and his research team used an AI program to turn sensor data into images. Like the previous analogy, the artificial intelligence’s images look like shadows contrasted against a grey background. 

READ: Facial recognition spots late-night gamers

It compiles those pictures into a video that moves in real-time. The researchers show that their algorithm can illustrate a user’s gestures, such as swiping, hovering, and scrolling. They conducted three tests to prove it.

1. They seated a dummy before the device while different hands moved toward the screen. A human hand pointed to the screen, and then they touched the monitor using a cardboard cutout.
2. The researchers showed a light sensor can provide real-time tracking by deploying a faster one. Consequently, the more powerful sensor enabled them to follow movements by one frame every 3.3 minutes.
3. The last test revealed that people are still at risk when watching videos like films and short clips. A light sensor can still capture gestures, even if the person has a whiteboard behind them.

“This work turns your device’s ambient light sensor and screen into a camera!” said Princeton professor Felix Heide. “As such, the authors highlight a privacy threat that affects a comprehensive class of devices and has been overlooked so far.”