Multiverse 2.357

What will facial recognition systems look like in five or ten years? Welcome to the first post in this whole biometrics series to consist entirely of idle speculation! After building up a basic description of the status quo, it’s time to extrapolate forward based on nothing more than intuition.

Some trends are obvious: increasing resolution of cameras, continued miniaturization of cameras, improvement of facial matching algorithms, the rise of machine learning (ML) systems to manage these tasks at scale, the explosion of affordable satellites, increasing availability of personal biometric data due to breaches from governments that they just cannot resist collecting and warehousing, and development of complementary technologies such as gait recognition.

If you haven’t already, go take a quick look at our recent post all about facial recognition systems. It covers the basics of the technology, the companies who want to bring these systems to market and is good backgrounder. Governments and corporations are developing this technology and are eager to deploy it, so get ready.

Amazon seems poised to deploy commercial versions for use by governments, and even build out networks of consumer products that interact with local and national law enforcement agencies. This comes with benefits in terms of fighting crime, but at the expense of personal privacy in open societies and of basic freedoms in more authoritarian ones.

China, on the other hand, seems to be content to develop this technology solely for governmental use; both internally and to sell to eager governments around the world. It’s estimated that China will have 570 million public cameras deployed by 2021, and already has more than any other country.

So what do these systems look like in 2025? I think a lot depends on how the facial recognition software systems evolve. I suspect we end up with a few different kinds of systems. We’ll definitely have competing software stacks, with China, US and proprietary systems from corporate giants having the inside track to becoming the main competitors.

I’m really talking about these systems becoming differentiated based on the hardware profiles where they’re used. For access control, you get high resolution close up images on cameras and that works wonderfully. You need more robust and varied systems for tracking people in crowded public spaces. Some estimates suggest there might be 500,000 cameras in London, although probably less than 50,000 of them are operated by the government, we’ll undoubtedly see networks deployed that can take advantage of the huge installed base of CCTV cameras. These are on wired and often wireless networks and need to transmit data to a remote location for processing.

The kinds of networks we should expect from giant corporations like Amazon will surely utilize much more advanced hardware. These are likely to be devices primarily intended for consumer markets with an eye toward integration with government networks. The physical cameras are certainly going to be better, but not prohibitively expensive, as consumers will be deploying a lot of these – perhaps the majority. Like the legacy networks of cameras, these systems will be designed to do all processing in centralized hubs.

China is quite likely in this author’s estimation, to develop systems that have very different characteristics. The hardware is intended for sale to governments and institutions. They have done early work aimed at integrating facial recognition systems with gait analysis software. When these two factors are considered in light of their goal of tracking citizens without regard to how intrusive that tracking is, it becomes clear that they will likely develop something unique.

If a country is committed to invest in deploying technology at scale to surveil their enormous population centers, it makes sense to design effiecient and effective new systems instead of patching up networks of legacy sensors. Instead of simply deploying cameras, why not have a data collection device that also has capabilities to thwart attempts to hide from the cameras – infrared and x-ray technologies come to mind, so mitigate challenges with lighting and people trying to provide deceptive views of their faces.

The software mix is likely to get varied and interesting as well, depending on the context it is designed to operate in. It can operate in phases, in order to work in crowded public spaces. Initially a scanning software can survey large crowd and identify the people, bicycles, dogs, trees, and other objects. The humans are tagged and the facial recognition and gait analysis softwares begin to process them further. Once identified, ongoing tracking can be accomplished by a complementary satellite network until they go inside buildings, enter a subway, etc. Special environments may require assistance from drones.

On the other hand, situations where cameras will always get close, high resolution imagery lend themselves to a different software stack. In cars, for example, a facial scan can be used for authorization, but there are other valuable data to be had in this contained physical space. The same hardware used to recognize a passenger’s face can be used for tracking pupil movement as well. This is useful data if the subject is navigating or has some limited control over vehicle operation. It’s also useful if advertisements are being displayed. The cynic in me wants to suggest that perhaps taxi operators will augment their income by showing targeted advertising to riders.

The identification of an individual face as matching one in a database is going to remain the heavy lifting that needs to be done in expensive data processing centers It is also likely to remain reliant on the fairly unreliable 2D facial data from images being matched up with 2D images in databases for some time to come. But if a massive amount of new hardware is being deployed, there’s no reason these multi-purpose cameras with other sensors can’t also have some very affordable processing power. China is definitely the place to watch for serious innovation in this space. Amazon and others like Google will continue to serve the consumer and commercial markets, but integration and cost containment will likely be the driving forces in the development of their systems.

I’d like to end this bit of unfounded speculation with an uplifting bit of optimism, but in fact the picture painted here is one of a dystopian world where multiple entities are going to surveil and track us every time we venture outdoors. If we hope to retain freedom and privacy, we’ll need some high tech camoflauge technologies to help us do so. I’ll happily take a crack at envisioning what that might look like in a future post, so stay tuned!