Computer vision in gyms to maximise benefits of your training - a good idea?

[Amarone2]

Is it possible to get feedback from members on whether the use of computer vision at your gym would be a good tool, or if it's just fluff solving problems that don't exist?

Imagine you have an app on your phone (could be a new app and/or an app you already use), and the gym has several cameras set up with AI. Cameras already exist at many gyms, but you might need some that capture really good angles and record the actual training in HD.

The cameras can be hidden in mirrors or placed in more fun/colourful boxes instead of surveillance cameras, so they look "friendly" and not like "big brother is watching you."

Possible use cases:

• The system recognizes you as a paying user of the app when you arrive. Only your face appears in your training videos.
• Automate logging all training set by set in real-time (exercise/weight/number of sets/number of reps, even exercises without weights and cardio). If desired, the app can vibrate when it's time for the next set based on the program you're following.
• The app can flag good form/poor form/dangerous form (e.g., heavy deadlifts with a rounded back or "kettlebell swings") and even ask if you really want to log that last cheat rep as a fully completed rep and move on to next week's weight, or if you want to redo this session.
• Video clips of sets are sent to the app, especially those that stand out as "bad" or "good." Users can then upload them to social media whenever they want.
• Based on form/endurance (which builds on speed, lifting angles, and form), estimate your 1RM max based on, for example, a heavy 3/5 rep set if you don't have a spotter.
• Overall assessment of the workout's quality based on you and your form.
• Program recommendations based on you, your physique, and your goals.
• Special images can be taken at regular intervals to show how your physique develops over time. With the right lighting/image quality, the system can likely see the progress that the human eye can't. These images might require a private space at the gym if people are shy or prefer to bare their torso.
• It’s possible that this could link to recommendations around diet/supplements over time.
• Reminders/motivation from the app to let you know it’s time to train.
• Local/national "competitions" between app users. The gym could have a large screen showing "Lift of the Month" or "Most Progress," etc., for those who choose to participate, and users could win, for example, x months of free training.
• Other use cases – feel free to jot down any thoughts you have :)

Thanks!

sorry just testing to see if the post got published or if my pc is acting up.

But AI couldn’t do all those things. Good form/poor form/dangerous form is subjective depending on what you’re doing. A rounded back in a deadlift isn’t intrinsically ‘poor’ form. Judging a 1RM involves skill and knowledge of how a lifter moves. Some people will lift slower than others, have different sticking points, different weaknesses.

IME AI is good at generating plausible nonsense which is what this sounds like TBH. I’m sure there will be a market for it, but it will be to the detriment of the quality of the input people get.

"How would you like multiple camera angles of you in the gym to be uploaded to some AI server somewhere and used for god knows what?"

True, that all comes back to how the neural network behind the model is trained. Shite in = shite out, meaning if it's trained on too little data and supervised by people whp don't know what they're doing then for sure it'll be bad.

There can also be a manual override in the app in the spirit of "I'm an experienced powerlifter and know what the heck I'm doing, don't flag lifts like this to me as poor form going forward."

For the 1RM I think there are nuances as well, if the app is targeted at competing powerlifters then it may require a lot of skill. If it's there to give an average gym-goer an indication of their 1RM, I think the positioning is quite different.

RE people's lifting speed - it's a learinng system so it will first need to create a 'normal view' of you as an individual and there determine strain from that.

A very valid concern and there needs to be 100% transparency on the fact its designed as an ethical system following GDPR and other privacy and cyber sec regulations and that people who opt out will not appear in any footage.

You do realise any commercial gym in the UK today is already packed with CCTV? As is any public environment in the UK...

-------

GDPR:

"[...] any information relating to an identified or identifiable natural person (‘data subject’); an identifiable natural person is one who can be identified, directly or indirectly, in particular by reference to an identifier such as a name, an identification number, location data, an online identifier or to one or more factors specific to the physical, physiological, genetic, mental, economic, cultural or social identity of that natural person

Yes, a video in the gym that contains imagery of your face is generally considered personal data under the GDPR. Your facial image is considered a biometric data and can allow for your identification, making it personal data. Recital 51 of the GDPR clarifies that photographs can be considered personal data, especially if they allow for identification.

If your face gets blurred, does that change anything?
Yes, if your face is effectively blurred in a video, it can change whether the imagery is considered personal data.

• Anonymization: If the blurring is done in such a way that you can no longer be identified, directly or indirectly, by any means reasonably likely to be used, then the data may be considered anonymized. Truly anonymized data falls outside the scope of the GDPR. The blurring must be irreversible to be considered proper anonymization."

--

Have you ever asked your gym what they do with their CCTV footage? If they save a porton of it, how is your PII handled?

Quick commentary on the proliferation of CCTV in London as an example:

And BTW, gyms are already rigged up with CCTV so that's not really an argument as far as I'm concerned.

It's believed that London has just under a million CCTV cameras in operation. In terms of government-operated cameras in London, per ChatGPT:

• Government/Public Cameras: The London boroughs, Transport for London (TfL), and the Metropolitan Police operate around 63,449 cameras. The Metropolitan Police themselves operate about 3,000 cameras attached to buildings and over 24,000 body-worn cameras. TfL oversees over 15,500 cameras across its transport network.
• Likelihood of Capture: It's estimated that the average person in London could be caught on CCTV up to 70 to 100 times per day.
• Most Surveilled Boroughs: In terms of cameras per square mile, the City of London has the highest density, followed by the London Borough of Hackney. However, Hackney has the highest overall number of CCTV cameras.
• London Underground: The London Underground network has over 15,516 CCTV cameras in operation.

Sounds to me like London itself is already rigged like one giant BB house!

With that said, what's the big deal if a gym was to add a capability for their members to automate parts of the work they do to track and improve their workout regimen, if it's done in a 100% transparent way?

Yes gyms have their own CCTV. But the (often shockingly low-quality) video from the cameras are typically stored on-site. They might have selectable streams that go off-site for remote monitoring but it won't be stored there. There is a reason for this.

Which leads us to the next issue with your exciting "Let's take something that already exists, sprinkle some AI pixie dust on it, and profit!!??!1!?!" business idea. You've got two choices - either process the video on-site, or ship it off to the cloud and process it there. If you process it on-site then you'll have to purchase, install and maintain a fair-sized chunk of hardware in every gym which will then be subject to theft, damage etc. Not a great idea. So you ship it off-site. You'll need multiple cameras taking images from multiple angles, and they'll need to be fairly high-quality. That's a lot of data that will need to be uploaded from the gym to the cloud. A lot. That's the reason why the CCTV is stored locally.

Tell me, how fast do you think the typical gym's Internet connection is when uploading? Ask Chat GPT to explain to you why broadband connections typically have a much slower upload speed than download.

I don't think you'd be able to make this GDPR compliant. Facial recognition has stricter controls and you're effectively collecting biometric information, which could fall into the category of medical information, which needs stricter controls and security. To cover the costs + hardware, the app price would have to be high so I don't think you'd get many subscribers. For £20-30 a month, you're close to paying a PT.

There is no way I'd use a gym with such a system in place anyway. I suspect most people don't care, but I don't think many would pay for the app.

May I suggest you tone down your adversarial and condescending tone. You have no idea who I am or what my motivations are so kindly refrain from further stupid remarks about profiteering.In relation to the actual questions I asked don't understand what you point is, if you have one... I take it this is definitly not linked to any current pain point of yours and that's fine.

Thanks for input to the technical solution though. I wouldn't target "Joe's garage" and tbh if a modern gym chain can't bother with getting a modern reciprocal fibre broadband connection I don't think this technology is of much interest anyway. Even if I had 20 concurrent HD streams using H.265, you can add that up, it's not that much on a modern fibre connection.

HW for neural networks becomes quite expensive quickly and it would be less costly to process it in the cloud or perhaps through a mix of AI directly on the cameras and cloud.

Thanks for your last comment, that's really the type of comments I was looking for with this thread. Need to ensure that the presence of such an app wouldn't turn people off whether they choose to use it or not.

Good point - You're absolutely right about the GDPR compliance aspect being stricter for biometric info and this is a point to figure out. An iphone stores the faical data of its owner only on the phone so Apple goes around it that way. This may have to be replaced by tags in a wrist band or general cues about a given member's look that day.

This app would not get into the territory of processing medical information, there needs to be a clear line in that it will not detect signs of fatigue and what have you beyond form. It will not monitor vital signs etc.

It wouldn't be possible to price an app like this anywhere near 20/month so that's completely out of the question. I'm amazed though that you through this chat have been able to infer such detailed price points.

Urgh, not for me. I am old and tight though.

biogym/technogym ready exist for circuits.
anything else needs instruction/supervision by a well trained human.

My day job is managing a network that carries around 80Gbps of video streaming traffic. That includes multiple gyms and a shit load of CCTV. How about you?

And do you think quoting ChatGPT helps you to present yourself as someone who knows what they're doing, or that it might in fact make you look like a dilettante?

But ok, you'll ignore any gym that doesn't have a 1Gbps+ symmetric fibre connection. That means that your target market will be the big chains who will have good network connections to run their PoS etc. And you'll be asking them to reserve a significant chunk of their bandwidth for a product that will be competing directly against the profit stream they get from their own PTs.

Well, you've convinced me. I can't see how you can possibly fail. You go for it champ!

Thanks for commenting.

Thanks for your perspective. Nothing stopping humans from instructing as I see it, don't see these as necessarily mutually exclusive but you're right, some element of 'knowing' is needed before going off on your own.

Not sure what your problem is but I hope you're a kinder and more positive person in real-life. You clearly don't have an interest in discussing the use cases this thread is actually about or knowing whether I fail or succeed (which isn't what this thread is about). You seem to be angry and wanting to insult me. I now understand you work with data comms and that it's important to you to come across as an expert at that (which I don't doubt you are) and that's great but that's not the aim of this thread.

Also, realise getting a novel tech service up and running is a long process and will require one or a few testbeds as pilots to run for X months or even 1+ year. By the time it's matured into a commercial concept, fibre broadband in the UK will have moved on as it keeps expanding every day. I assume you as a comms professional are aware of OpenReach's April 1 announcement. You'll keep seeing announcements about new fibre availability in UK regularly.

And what's to say one couldn't start in say London, where fibre BB is prolific for businesses, and then expand from there. Nobody's talking about securing business from 500 gyms tomorrow.

All: Given the few replies I've received plus the overall lack of engagement here on mumsnet, it's an early indication that this isn't being a particularly interesting concept for mumsnet's demographics. I've received more positive feedback from younger audiences and audiences outside of the UK.

I’d suggest if you want more engagement, you could reflect on how this whole exchange went. And if you want feedback from people how you communicate with them. I gave up as you seemed to think you knew it all anyway, or if not would get it from chatGPT, plus a dose of attitude about that. A good researcher works to create a space in which people say what they think, and you’ll then get the gems you’re after amongst stuff you think is ‘wrong’.

I’m sure you’ll tell me off for this, and why I’m wrong, and maybe give me chatGPT’s guide to research.

My teenage son is highly into fitness, weights, gym, and would be interested in seeing outputs for recommendations, if not convinced by OP's product. The competition element would attract him. Last year, a local young person managed to kill himself with weights on gym equipment. I am trying to imagine if this system could stop someone from "doing something" stupid with the heavy weights.

Who wants to upload social media videos that are only showing their face grunting & sweating thru a workout? Didn't think that one thru.

For this objective:

• Overall assessment of the workout's quality based on you and your form.

Sounds impractical, a huge use of environmentally destructive AI that won't be as good as a human could do with information on personal history, workout routine & body composition that the user can more directly collect and provide (so, that objective would be the wrong way to use AI). Also, fitness is mostly a psycho social outcome and the AI can't capture those aspects. Whereas, a Good Personal trainer service helps people manage competing demands in their life that prevent them getting to the gym (or any exercise) or eating the best foods or getting enough sleep, would be far less environmentally destructive and 10x more motivating & effective ... some gyms have limited PT services included in membership cost, so wouldn't be cost effective to add on the App that OP describes (by the time OP pays for the AI processing & servers & cameras & admin etc)

My (Olympic size rec centre) pool can even afford SwimTag never mind the fancy sort of service OP describes. I would love to have SwimTag again (!!)

my exFlatmate had been very obese & trimmed down for years afterwards. He rigorously did exactly what his PT told him (eating & workout)... thing is, he was fussy about his PT, who had to be someone highly supportive on a personal level, someone flatmate liked. OP's AI just won't do that.

An observation. You haven’t indicated a target price for this, but I’m going to guess that it won’t be particularly cheap because of the hardware cost per gym that will, inevitably, have to be amortised over the relatively small proportion of the users of the gym that sign up. I am sure that younger audiences will express more interest, but I expect they have less disposable income. Have you considered the Peloton approach - sell expensive hardware for home use to cash rich, time poor, competitive cycling fanatics, and charge a subscription as well?

As a fifty something gym goer that does use free weights, the proposition you are pitching doesn’t seem attractive compared to my current approach of paying a PT for a couple of sessions a week. I want a tailored, proactive approach from someone who understands me, my body and my objectives and that I can have real time dialogue with.

I’ve nothing technical to add to this. A lot of it reads as if it was tarted up using AI.
It is a textbook case in how not to engage with your audience. I certainly won’t be passing on any useful comments to this OP!

I agree with a previous poster about your attitude. You seem to want sycophantism, but you're asking a demographic that actually runs tech start ups and has the money for such apps. How many of your enthusiastic young people have either? Pelaton users skew older, wealthy and female just like MN, so maybe check your ageism/sexism before dismissing us.

@lljkkmy instagram feed is basically a stream of videos of fellow lifters showing their training sessions. Powerlifters frequently film their training so they can send to their coach and/or review their own technique. And then we all share and watch each others. So that bit is definitely a thing

" Only your face appears in your training videos."

Is that how your Instagram feed is, Ray, you only see the faces?

@lljkki thought the OP meant only your face, not the faces of others. So rather than the user manually blurring out or covering other people’s faces, it would recognise your face so leave yours in and remove others. The video clips are referred to later as ‘of sets’, so they would need to be of the whole body.