Gait recognition technology ( = identifying a person by the way they walk) is getting good, and I find this trend unnerving. 

Current context: It's already deployed in China by a company called Watrix, which claims 94% accuracy up to 50 meters away from any angle. Japan, the US, and the UK have also been working on gait recognition, though I doubt it's being seriously deployed in the same way as in China. 

And it will only get better. Here is a recent paper out of Zhejiang University in which they generate synthetic data for training gait recognition systems (import AI 252 summary here). 

What makes me uneasy is the supposed robustness of gait recognition technology: 

“You don’t need people’s cooperation for us to be able to recognize their identity,” Huang said in an interview in his Beijing office. “Gait analysis can’t be fooled by simply limping, walking with splayed feet or hunching over, because we’re analyzing all the features of an entire body.” ~ AP News article [Note Huang is the CEO of Watrix]

Main question: In the future, if my local grocery store installs cameras and uses facial recognition technology to track my purchases and show me targeted ads, I can probably prevent this by wearing a face covering. What's the analogy here for gait recognition—how do I make myself confusing and/or undetectable to such systems? 

Some random ideas I had: 

  • Wear ankle weights, and vary the amount of wait on each leg
  • Wear strange shoes (e.g. high heels)
  • Wear a fat suit 
  • Wear a very heavy backpack
  • Walk in silly ways (limping is said to not work, but what about side stepping, etc) 
  • Physical therapy to be able to switch walking styles deliberately 

I'm thinking about mostly with the "lazy surveillance" use case in mind, which is what I consider to be the most likely outcome in the western world. For example, I am subjected to "lazy surveillance" by Facebook/Google/other tech companies, but I can circumvent this by using a VPN, incognito window, different accounts / locations, etc. Hence the motivating example of the grocery store—a low stakes but realistic future scenario. 

New Answer
New Comment

4 Answers sorted by

jimrandomh

180

Machine learning metrics are tricky; if you don't know what they mean, they tend to sound impressive, when they really aren't. 94% accuracy is actually terrible, to the point where I would call this a scam if it looked like it was being marketed B2B. Consider: If some company has a database with a million people in it, and this technology rules out 94% of possibilities, then this puts you in a group of 60,000 people. This is about the same accuracy is they'd get if they just measured your height, and ruled out everyone more than an inch shorter or taller than you. (In fact, I'd put pretty high odds on this being exactly what the "gait recognition" neural network is actually doing.) So it might work as a cross-check in combination with some other tracking technology (eg your phone's MAC address), but if that happens, it's the other tracking technology you should be focusing on.

As an ML practitioner, that's not what I'd mean if I said "94% accurate". I would mean that the label was correct 94% of the time. This is very much affected by the size of the db -- that is probably why the use a weaselly phrase "can reach" -- "The average recognition rate can reach 94.1%"  says the Watrix link.

This is a good point concerning current gait recognition technology. However, I don't doubt it will improve. On longer timescales, this should happen naturally as compute gets cheaper and more data gets collected. On shorter timescales, this can be accelerated using techniques such as synthetic data generation. 

Perhaps there is a natural limit to gait recognition, if it turns out that people can't be uniquely identified from their gait, even in the limit of perfect data. But if there isn't, then in 10 years, "94%" will turn into "99.999%", or whatever... (read more)

ChristianKl

140

Given that most shoppers in Western grocery stores currently carry mobile phones that tell anyone who's nearby about their identity. There's little need to use anything else to identity shoppers. 

I had an experience where a friend changed her body tension patterns enough that I didn't recognize her after she was a week on vacation.

Physical therapy to be able to switch walking styles deliberately might be problematic because having the ability to move in a very conscious way is likely to produce tells. Whether or not the gait recognition algorithm can pick up on them however is another question and probably not at the level of only giving 94% accuracy. 

Japan, the US, and the UK have also been working on gait recognition, though I doubt it's being seriously deployed in the same way as in China. 

A decade ago the London was of the mindset that the limiting factor for their surveilance was disc space. 

I would be surprised if the Five Eyes don't have gait recognition.

This video on how Usonian CIA operatives disguise themselves claims a simple piece of gravel in your shoe will change how you walk completely: https://youtu.be/JASUsVY5YJ8

2ChristianKl
It will change some patterns of how you walk. Likely, those visible to most people. There are other things that it likely won't change. If you for example take the gait pattern of police officers to hold their hands more laterally then the average person which comes from walking around a lot with a gun on your belt I would be surprised if that pattern simply disappears when you put a gravel in someone's shoe. I'm uncertain to what extend a gravel in someone's shoe would affect my recognition if the gravel is just there for the interaction and not walking around a month with the gravel. The gravel changes muscle activiation but it doesn't do anything to change the fascia patterns in a short amount of time. It's worth noting that gait patterns are more central to me perception of other people then most people. One example for that was a workshop at a dance festival where the dance instructor was demostrating how he changed the way he was doing a dance step and then asked what he changed. I was able to pick it up, but it seems like I was the first person to do so and the dance instructor asked whether I have been at the workshop before (he did the same workshop at many different festivals).   One interesting aspect of the video is the claim that while women can manage to pass as men, men can't manage to pass as women. From what I know about transpeople, there are transpeople who manage to present as the opposite sex in both directions. It might take more work for a male to present as female but it's possible. 
1methree
Usonian = from the USA, which doesn't offend the sensibilities of the rest of the countries in the continent of America

Euglossine

20

The big problem I see with all of the methods you describe to avoid matching is that they are one-offs. A mask obscures your identity; you are attempting to make your face indistinguishable from anyone else's. But alterations like (most of ) the ones you describe just create a separate identity -- more akin to plastic surgery. This is great if you are trying to avoid linking yourself to a former identity, but not as useful for trips to the grocery store, where this new ID will be tracked and monetized.  

My suggestion would be to wear something that hides your gait, rather than alters it. A burqa might work. Or ride in the motorized cart. 

Regarding the goal: Personally, I find the tracking by the grocery store helpful enough (given the incentives they set up) that this is among my areas of least concern for tracking. I appreciate, for example, having novel grocery products that I might like suggested. This is different than for, say, televisions, where I discover that I am artificially dissatisfied by learning about newer features. A key differentiator is that I am constantly buying new food, so there is no pressure to buy something vs. remain satisfied with something I already own. Of course, other areas where gait tracking would come into play might provide a different value profile for me.