New camera design
The new camera design works on the principle that most devices don’t need to see as much detail as regular cameras provide to complete their tasks. Most utilise only a narrow scope of visual signals, such as colour or pattern recognition, to let them recognise objects, navigate through space and assess changes in their environment.
Therefore, visual information can be drastically reduced, distorted, or scrambled in such a way that it provides everything the robot needs, while preserving the anonymity of its human owners.
Most researchers around the world tackling the privacy problem have focused on scrambling and encoding the images digitally within the device’s in-built computer, which still leaves the original images vulnerable to hackers in a cyberattack.
The new Australian breakthrough redesigns the optics and analogue electronics of the camera itself, so that it only collects the specific visual information necessary to perform a certain task.
For example, it can establish whether it has previously visited a particular location by steering a single mirror, one pixel wide, in a random pattern around the scene.
The camera’s circuits collect statistics on the relative lightness and darkness of various points, converting it to a unique “fingerprint” that is sent to the device’s computer for reading. Full images are never received by the camera and therefore can never be leaked, misused or hacked.
“So much of the approach to the technology today is about upgrading digital capacity so we can collect more and more data,” says Dr Don Dansereau, who led the project at the Australian Centre for Robotics.
“But when it comes to privacy, there is a strong argument for returning to fixed analogue components that put hard limits on the amount of data that can be captured.”
Commercial potential
If the new camera design lives up to its promise, commercial returns could be great. It could play a big part in the growth of the so-called internet of things, especially as the growing global focus on consumers’ right to privacy, embodied in legislation such as the European Union’s AI Act, imposes increasing compliance obligations on technology providers and importers.
The researchers are currently looking for industry partners to explore manufacturing at scale, and hope to have their cameras in the market within the next two to three years.
Professor Niko Suenderhauf, deputy director of the QCR, who advised on the project, says he hopes to see the technology used in many applications, particularly for environments where privacy and security are a concern such as schools, hospitals, factories, warehouses and airports.
The team also thinks autonomous flight applications, such as delivery drones, will be a potential market for their privacy-preserving cameras.
The team has issued a challenge to the research community to try to hack its technology. In the meantime, it is fine-tuning the camera design for different tasks, including navigating a mobile device through 3D space, and constructing physical prototypes to demonstrate its approach in practice.