AI surveillance camera tells you what it sees when it recognises you

shutterstock_260530250It’s been reported that New York City’s police department can tap into about 6000 street cameras, two-thirds of which are privately owned, with another 7000 in public housing and more than 4000 in the city’s subway stations.

They’re not easy to count with precision.

But it’s likely that many major cities are on par, if not even more densely stuffed with surveillance cameras.

That means that many of us can count on having our image captured multiple times per day.

In fact, NGO Big Brother Watch has estimated that the average Londoner is recorded on CCTV over 300 times every day.

Wouldn’t it be interesting to know what those cameras are thinking as they’re capturing our images and movements?

That sounds odd, I know, to credit a CCTV camera with “thinking.”

But as it is, there’s at least one company, AISight, working on a CCTV network that’s outfitted with artificial intelligence (AI) and which Boston – stymied as it was in its efforts to identify the Boston Marathon bombing suspects through facial recognition – has reportedly contacted to explore ways to spot crimes before they even happen.

So yes, there are “sentient” CCTV cameras that learn and that can, like AISight’s network, spot anomalous activities – e.g., why is that car backing up the ramp into the garage? And why has person X visited the rear of the building multiple times? – that could be precursors to crimes.

Back to the question: wouldn’t you like to know how such an AI surveillance system interprets you and your movements?

There are two masters students at New York University who have invented a surveillance camera that’s designed to answer the question, quite literally, reading aloud in real-time its interpretation of what it sees when it captures a facial image.

As Motherboard reports, New York University masters students Ross Goodwin and Gene Han, who are researching artificial intelligence and machine learning, invented the talking surveillance camera.

They’re calling their DIY, interactive art project a “sentient surveillance camera.”

Goodwin told Motherboard’s Brian Merchant that the point is to raise awareness about what he called the “omnipresence of surveillance equipment”, as well as the current state of technological advancement with AI.

We wanted to create an entity with its own sense of social awareness, its own eyes, and an ability to communicate with humans, albeit with some glitchiness that underscores the limitations of the current technology.

Motherboard shot a video of the camera in action in Brooklyn, New York. It shows the limitations to which Goodwin referred: it captures images, then “talks” about its findings, via Apple’s text-to-speech utility, in a surreal, garbled stream, little of which was decipherable when I viewed it.

But hidden within the stream of machine speak were nuggets that raised eyebrows among the passersby who took part in the experiment: one woman was correctly identified as a traveler (she was from London), and the other was tagged as having something to do with higher education (she’d recently graduated from college).

Goodwin said that the panning/tilting/surveilling camera is constantly moving and scanning for human faces.

When it recognizes a face, it uses Haar Cascade detection, which Motherboard’s Marchant describes as “a machine learning approach where a cascade function is trained from positive and negative images.”

The camera zooms in on the face, Goodwin says, and sends an image of the face and its surroundings to its creators’ server, which then uses convolutional neural networks to extract concept words from the scene, after which…

Those words are expanded into sentences and paragraphs using related words from a lexical relations database. The paragraphs are then automatically read aloud using Apple's text-to-speech utility.

Marchant notes that some people who volunteered to be scanned were delighted at what the sentient machine “knew” about them, but some were horrified.

Even if it only got a word or two right among all the gibberish, the fact that the sentient surveillance camera knows anything at all about us is a far more concrete demonstration of the power of AI-assisted surveillance technology than any of us are likely to confront as we walk beneath the normally silent watchers.

What’s the worst that can be done? Well, we already know that facial recognition technology can be used to tailor adverts to us.

The UK’s largest supermarket chain, Tesco, moved to install facial recognition technology in its petrol station forecourts a few years ago, and it’s already being used to potentially sell us beauty products.

Is there anything you’d like to say to these sentient surveillance cameras?

Share it with us in the comments section below, and we’ll pass it on next time we pass under one that’s willing to engage us in conversation.

Image of automated surveillance camera courtesy of