Matt Webb has posted some thoughts he's pulled together to introduce a discussion he's going to be taking part in on security and privacy. I was particularly tickled by this anecdote:
Hoxton Analytics supply, for their clients, pedestrian footfall intelligence. They count the number of people walking in and out of your shop.
Note the precise thing they're being asked to count: footfall. That's going to be important in a bit…
[Various reasons why facial recognition won't fly, and other techniques have their shortcomings too.]
Hoxton takes a different approach. They have cameras right down low on the floor, and they use machine learning - on the device - to recognise shoes.
It's crazy accurate. 95% accurate. It can also count group sizes, and whether people are going in or out. So it can do capacity.
Bingo! All you need to be measuring is feet, falling.
It also doesn't store personally identifiable information so it's good in Europe.
But get this. Because they've built this solution, it means they can also use it in public places. So you can point the camera out of the window and see how many people are walking past, versus how many people are walking in. This is the holy grail, like a conversion funnel, like Google Analytics, but for physical retail. And they've got there by considering privacy not as a product constraint, but as a product feature.
There's a little part of me that wonders whether the roots of this approach came when someone in the team was sitting racking their brain over how to solve this problem they'd been presented with and suddenly the notion "If all they want us to count is footfall, all we need to worry about is counting feet!" popped into their mind. AI for counting shoes: why not? No need to build a world of data about whose feet, or build up a deep, detailed picture of what else the owners of those feet bought in the last 30 days or whatever. Just count (potential) customers moving around your premises, be grateful it's more accurate than the data you currently have, and work from there.