When I put my amazfit helios strap to record a workout of type Strength, it automatically tries to detect the movements I'm doing and figure out what muscles I'm working out. It actually works, not 100%, but it works. It tracks my pull ups, my push ups, my dips, really good. I was wondering how they do it but didn't look more into it.
I'm sure other watches/trackers do that as well. It's really impressive what we can do with just a watch.
Not OP, but yes - look up Motra (formerly Train Fitness). Been using it for a year in the gym and it feels like magic!
Accuracy is perhaps 60-80% for the kinds of exercises I do - doesn’t sound like a lot but still saves time (and more importantly, reduces the need to fiddle with screens during a workout). Sensible defaults + great UX helps in cases where the auto-detection fails.
I think it’s helpful to conceptualize it in statistics terms. Second-order inference will be inherently inaccurate with a limited single input, but it’s hard to convince people to wear two smart watches on each arm.
My garmin 8 tracks chin ups very well to without any additional hardware. Its like 10% deviation from the total number.
Funnily, my fenix 6 I had before was more like 1% deviation. It tracked chin ups REALLY well.
besides this: its very scary what the 8 is tracking and how good it actually is. I had a covid case this year and the health metrics were all over the place and the warning lights lit, including skin temperature.
On the other hand, the "in range" value in that screenshot was a blood oxygen of 96%?
I'm not saying these devices don't work, but the values displayed are all relative and should be taken with a grain of salt.
Even the best pulse oximeters used in hospital care have this problem. Small blood vessels constrict for any reason from being cold to stress or even just standing up.
The accelerometers are the star of the show, not the other sensors.
I see a lot of these comments namedropping brands and models, but that can be misleading for some potential customers. They all still suck pretty bad in some common and some slightly different ways. I would not even depend on any smart watch or smart ring to tell me if I have sleep apnea.
The pulse ox does not work. Wrist-based pulse ox sensors cannot work reliably. Heart rate and motion work. On my Fenix 6 I tested the respiration and it doesn't work. Skin temp works, but it's not a very useful metric (unless you're in the water, in which case it very accurately measures water temperature.)
That said their calorie metrics work excellently in my experience. Not perfect, but good enough to be better than anything I could do by hand, and good enough to actually track calorie usage. The calorie metrics use both accelerometer and heart rate as input though.
Sleep pulseox on my fenix 7s is garbage. I assume, because it frequently reported very low values for me while sleeping.
But in the measurement mode where you touch the dial and sit still for a minute, I've found it to be 100% concordant with a finger monitor used at the same time. over multiple reading on multiple days.
This was in a high-altitude environment where I was having trouble, so it's not a case of "both report 99%, so they're in agreement". I think I was down to 90% or some such low number.
> Sleep pulseox on my fenix 7s is garbage. I assume, because it frequently reported very low values for me while sleeping.
> But in the measurement mode where you touch the dial and sit still for a minute
Im pretty sure while sleeping it realized you are "still" and does the pulseox measurement the exact same way as the manual "dont move for 30 seconds".
That aside, pulse oximetry is the least interesting measurement to me.
I find HRV very interesting. When iam doing not well or my body is fighting an infcetion or something it definetely shows that in my HRV graph.
Skintemperature is new for me. The only time it warned me about that was the first night fighting covid. Sample size 1 but very accurate.
My Garmin also does this. What's most impressive is it seems to also detect when I'm doing one sided exercises. If I do six pistol squats on one leg and then six on the other it somehow figures that out and counts it correctly as six reps rather than twelve.
On the other hand, there are some movements it really struggles with recognising, like pseudo-planche pushups. It doesn't even register the reps as happening at all.
If you own an iPhone, go to Health app and scroll through your mobility stats. You'll find step length, double support time, walking asymmetry, etc.
Apple released a document in 2022 where they documented these features[1], along with details on how they work. I guess next versions will have more accuracy.
Similarly Garmin run watches [1] can tell ground contact time of each foot, your vertical oscillation and how much you slow down when you step down and speed up when you push off for each foot.
I think a chest HR monitor is also needed for some of those measurements
Used to be true, then Apple demonstrated that those metrics can be obtained using just a watch, so Garmin upped their game and got rid of the HR strap requirement.
And these metrics aren't even from the Watch; they're derived from the iPhone, which I assume is going to have a greater possibility of locations where it's stored on the body.
If you’re on Apple Watch, look up Motra (formerly Train Fitness). Been using it for a year in the gym and it feels like magic!
Accuracy is perhaps 60-80% for the kinds of exercises I do - doesn’t sound like a lot but still saves time (and more importantly, reduces the need to fiddle with a screen during workouts). Sensible defaults + great UX helps in cases where the auto-detection fails.
The product is for the elderly. Humans loose access to the higher pitches as we age. The effect is rather dramatic. There were once ultrasound products meant to selectively dissuade young people from hanging around areas they were unwanted.
>> The Mosquito device, whose high-frequency shriek is audible only to those under around 25, has been bought by police, local councils, shops, and even private home owners, to tackle concerns over groups of young people congregating and causing disruption.
I must report that this "under 25" is not very accurate, at least with 30-40 I sense _something_ not necessarily really hearing anything but there is something my brain notices and its very annoying, I once rotate such a device in a hotel. outdoor restaurant because it was super annoying, apparently the thought keeping children out of that area.
Guessing it's similar to the PWM issue. Some people are more sensitive than others, and 'under 25' is an approximation which is probably true a majority of the time.
I guess I'm unusual, because one of these devices used to drive me nuts by the entrance to a grocery store until I figured out why there was a high-pitched whine there. Or maybe the device was malfunctioning and working at a lower range. In any case, I'm definitely not under 25 and I think I can hear at least some of these devices.
Interesting. I live near a school, and my street is a conduit for police and emergency vehicles. Once every few days I'll be sitting in my backyard and hear a sudden deafening ultra-high pitch whine for about 10 seconds, imagine tinnitus on steroids. Can't tell where it comes from. I wonder if someone's got one of these things nearby.
My friend famously said "I can see that a TV is on by just looking at it" I think he meant he could hear if the TV was on or off because he could hear the ultrasound in the electronics.
Ah the good old days. I'm not sure if I would detect a CRT today, but I remember when I was younger that I could immediately tell when there was one running anywhere nearby. You can't really hear them, but you can hear them.
cool but also I'm a bit skeptical of ~whole body in the title. This article mentions tracking joints down to the ankles but I don't see how well it actually tracks the legs when they move independently of the upper body. What happens if you sit still and move one leg, or do a squat with your knees coming inward? Would it pick that up, or give you something closer to the squat it expects?
ive worked on a related problem, counting reps from a single chest sensor [1]. We covered 12 exercises and released the dataset on huggingface, but even avoiding false counts from movement between exercises took work. We also assumed we knew which exercise someone was doing.
anyways super curious what the acoustic signal adds when there's barely any chest or wrist movement.
Yea it would be really impressive if they could do this well, I've worked with wearables as well in the past, this seems like such an interesting solution
Acoustic signal is the whole innovation here I think. the movements shown in video at top of link aren't distinguishable from accelerometer alone.
The watch is basically sonar, so distance from sensor and not (necessarily) accelerometer action is the limiting factor. Which is to say, the ankle detection could have issue, but more because it's farther from the watch than because it's not attached to your torso.
Different frequencies are likely going to have different signatures for each movement and require an entirely independent training set. So agree that could work, but it's probably not easy to add arbitrarily many bands for a crowded gym
Not quite to the same level, but I've recently been working on something similar with my rowing app [1].
The most common use-case is to get rowing data directly from a PM5 or FTMS rower using Bluetooth, but if you don't have a smart rower it uses Apple Watch motion to detect your strokes / stroke rate.
It detects peaks on the pull, then ignores the recovery so it doesn't detect it as a second stroke.
It adapts to how hard you're pulling as you go, so it picks up your strokes within the first few without any setup.
Overall this works well for rowing, but obviously can't accurately work out distance / power etc.
Nice! I'm a Concept2 rower, what advantages does it have over the built-in software and ErgData app?
Also, you need to update your little rowing person animation at the top of the page; they're not hinging their hips enough and they're engaging their arms way too early. It's a silly little thing, but as someone who focuses on form it jumped out at me when I loaded the page.
Thanks for the feedback - I actually traced the form for the animation from the manual (pamphlet?) that comes with the RowErg but I hear what you’re saying.
The differentiator of this app initially is more about the focus on making rowing a habit (we also make Streaks, a well-known habit tracker), and deeper integration with things like Apple Health and Apple Watch.
We’ve only just released it recently, but further customization of workouts and the display is where we’re heading.
I did something similar about ten years ago for sculling using the a mobile phone gyro to detect the scull's surge with each stroke to get a stroke rate. The code was based on an algorithm described in a paper I found online, but coded from scratch. I was rather proud of that. GPS gave an approximate speed.
Such an interesting direction, especially for older adults. I hope I live to see the day when devices like this can genuinely help our aging parents stay active and exercise safely, even in remote areas where access to a qualified trainer is limited.
Nah, they will all be frail, some earlier, some later.
It will be more the age of robots. Just look at Asia
with that and you can see that they shape the future
there, not just China - really, Asia. It's a bit sad
to see how the USA and Europe decline in comparison
here.
Specifically, the left speaker transmits C-FMCW chirps sweeping from 18kHz to 21 kHz, while the right speaker transmits in the range of 21.5 kHz to 24.5 kHz. Each speaker transmitsC-FMCW chirps with = 600 samples per chirp. With a 50 kHz sampling rate, the sweep period for each chirpis 600/50000 = 0.012 sec (12 milliseconds). Assuming a speed of sound of 343 m/s at room temperature (20◦C),and following the formula provided in 2, the resolution of the EchoMotion sensing system is approximately 0.343cm (or 3.43 mm), and the sensing range is about 205.8 cm (approximately 2 m).
I bet it could technically detect breathing issues like sleep apnea but they shy away from any wording that would resemble a medical diagnosis, which would immediately trigger regulatory approvals like for ECG features.
In theory you can extract respiration from RSA and possibly even by looking closely at a chest-based EEG but if you want dependable results you want a strap that can feel the pull on your chest or a radar pointed at it or something.
When you attempt biofeedback in the 1970s paradigm (look at the biosignal in real time and extend your brain’s sensorimotor maps) as compared to the 2020s paradigm (sit with a $300 device for 3 minutes, get some score and say “gee whizz!” you learn right away that many biosensors are a joke.
Have you seen the video where the inaudible passage of planes are visible in the audio spectrogram, revealing a doppler shift which can be correlated with public ADS-B databases to determine the location of the recording?
So its taking accelerometer data from a watch, and using it to predict body position with a model trained on whole body motion as measured by ultrasound. neat.
No, from the article: “ It uses the built-in speaker and microphone on a commercial smartwatch to determine a person’s body position using a sonar-like sensing system.” The move to commercial smartwatches uses just their single speaker and mic, rather than their custom unit they built that had dual speakers and mics.
Hmm, so does that mean if an AI LLM is trained in everything kinesiology, human body movement, anatomy, physiology, and biomechanics using this device's capabilities, it can act as an ultra physical therapist or even go as far as a physician that will actually cure their patients just by "watching" a human walk. Pun intended :)
We could argue that the age of computing before about 2005 was the teenage years of the PC and before 2018-2020 was the teenage years for mobile devices.
Look at the rates of improvements since. For general computing they're marginal for the average person, maybe a tripling of performance every 15 years, which isn't nothing but it's also not life changing.
I.mean, follows the same principle of a Nintendo Switch/Wii controller? Games like Just Dance require certain movements and these are tracked with the Joycons
When I put my amazfit helios strap to record a workout of type Strength, it automatically tries to detect the movements I'm doing and figure out what muscles I'm working out. It actually works, not 100%, but it works. It tracks my pull ups, my push ups, my dips, really good. I was wondering how they do it but didn't look more into it.
I'm sure other watches/trackers do that as well. It's really impressive what we can do with just a watch.
I’ve been wanting an Apple Watch app that can do that and automatically track my lifting movements, but I haven’t found any yet. Do you know one?
Not OP, but yes - look up Motra (formerly Train Fitness). Been using it for a year in the gym and it feels like magic!
Accuracy is perhaps 60-80% for the kinds of exercises I do - doesn’t sound like a lot but still saves time (and more importantly, reduces the need to fiddle with screens during a workout). Sensible defaults + great UX helps in cases where the auto-detection fails.
I think it’s helpful to conceptualize it in statistics terms. Second-order inference will be inherently inaccurate with a limited single input, but it’s hard to convince people to wear two smart watches on each arm.
How about a phone in the pocket and a smartwatch though?
My garmin 8 tracks chin ups very well to without any additional hardware. Its like 10% deviation from the total number.
Funnily, my fenix 6 I had before was more like 1% deviation. It tracked chin ups REALLY well.
besides this: its very scary what the 8 is tracking and how good it actually is. I had a covid case this year and the health metrics were all over the place and the warning lights lit, including skin temperature.
This was the first night: https://i.imgur.com/ws1jTv4.jpeg
On the other hand, the "in range" value in that screenshot was a blood oxygen of 96%?
I'm not saying these devices don't work, but the values displayed are all relative and should be taken with a grain of salt.
Even the best pulse oximeters used in hospital care have this problem. Small blood vessels constrict for any reason from being cold to stress or even just standing up.
The accelerometers are the star of the show, not the other sensors.
I see a lot of these comments namedropping brands and models, but that can be misleading for some potential customers. They all still suck pretty bad in some common and some slightly different ways. I would not even depend on any smart watch or smart ring to tell me if I have sleep apnea.
The pulse ox does not work. Wrist-based pulse ox sensors cannot work reliably. Heart rate and motion work. On my Fenix 6 I tested the respiration and it doesn't work. Skin temp works, but it's not a very useful metric (unless you're in the water, in which case it very accurately measures water temperature.)
That said their calorie metrics work excellently in my experience. Not perfect, but good enough to be better than anything I could do by hand, and good enough to actually track calorie usage. The calorie metrics use both accelerometer and heart rate as input though.
Sleep pulseox on my fenix 7s is garbage. I assume, because it frequently reported very low values for me while sleeping.
But in the measurement mode where you touch the dial and sit still for a minute, I've found it to be 100% concordant with a finger monitor used at the same time. over multiple reading on multiple days.
This was in a high-altitude environment where I was having trouble, so it's not a case of "both report 99%, so they're in agreement". I think I was down to 90% or some such low number.
anecdata I suppose, but convincing enough for me.
> Sleep pulseox on my fenix 7s is garbage. I assume, because it frequently reported very low values for me while sleeping.
> But in the measurement mode where you touch the dial and sit still for a minute
Im pretty sure while sleeping it realized you are "still" and does the pulseox measurement the exact same way as the manual "dont move for 30 seconds".
That aside, pulse oximetry is the least interesting measurement to me. I find HRV very interesting. When iam doing not well or my body is fighting an infcetion or something it definetely shows that in my HRV graph.
Skintemperature is new for me. The only time it warned me about that was the first night fighting covid. Sample size 1 but very accurate.
My Garmin also does this. What's most impressive is it seems to also detect when I'm doing one sided exercises. If I do six pistol squats on one leg and then six on the other it somehow figures that out and counts it correctly as six reps rather than twelve.
On the other hand, there are some movements it really struggles with recognising, like pseudo-planche pushups. It doesn't even register the reps as happening at all.
If you own an iPhone, go to Health app and scroll through your mobility stats. You'll find step length, double support time, walking asymmetry, etc.
Apple released a document in 2022 where they documented these features[1], along with details on how they work. I guess next versions will have more accuracy.
[1]: https://www.apple.com/healthcare/docs/site/Measuring_Walking... (PDF)
Similarly Garmin run watches [1] can tell ground contact time of each foot, your vertical oscillation and how much you slow down when you step down and speed up when you push off for each foot.
[1] https://www.garmin.com/en-US/garmin-technology/running-scien...
[EDIT]: I think a chest HR monitor is also needed for some of those measurements
I think a chest HR monitor is also needed for some of those measurements
Used to be true, then Apple demonstrated that those metrics can be obtained using just a watch, so Garmin upped their game and got rid of the HR strap requirement.
And these metrics aren't even from the Watch; they're derived from the iPhone, which I assume is going to have a greater possibility of locations where it's stored on the body.
> stored on the body
or in, just for completeness' sake.
swallows-phones-George screwed the machine learning algorithm and shouldn't have been counted
If you’re on Apple Watch, look up Motra (formerly Train Fitness). Been using it for a year in the gym and it feels like magic!
Accuracy is perhaps 60-80% for the kinds of exercises I do - doesn’t sound like a lot but still saves time (and more importantly, reduces the need to fiddle with a screen during workouts). Sensible defaults + great UX helps in cases where the auto-detection fails.
(Not affiliated in any way, just a happy user)
Note: ultrasound for humans is not necessarily ultrasound for pets.
And not necessarily for humans in this case, it emits from 18kHz to 24.5kHz. Completely within hearing range for most mammals (https://en.wikipedia.org/wiki/Hearing_range)
The product is for the elderly. Humans loose access to the higher pitches as we age. The effect is rather dramatic. There were once ultrasound products meant to selectively dissuade young people from hanging around areas they were unwanted.
https://www.theguardian.com/uk/2007/mar/17/society.humanrigh...
>> The Mosquito device, whose high-frequency shriek is audible only to those under around 25, has been bought by police, local councils, shops, and even private home owners, to tackle concerns over groups of young people congregating and causing disruption.
[Insert appropriate Abe Simpson clip.]
I must report that this "under 25" is not very accurate, at least with 30-40 I sense _something_ not necessarily really hearing anything but there is something my brain notices and its very annoying, I once rotate such a device in a hotel. outdoor restaurant because it was super annoying, apparently the thought keeping children out of that area.
I'm in my 40s and I can hear those "under 25" sounds.
I wonder if there’s some built in assumption that most adults will have damaged hearing, left over from when manufacturing provided a lot of jobs?
But yeah. Even some laptop chargers seem to have this “keep me out of the area” function built in (accidentally, I guess) and I am not young.
Guessing it's similar to the PWM issue. Some people are more sensitive than others, and 'under 25' is an approximation which is probably true a majority of the time.
I guess I'm unusual, because one of these devices used to drive me nuts by the entrance to a grocery store until I figured out why there was a high-pitched whine there. Or maybe the device was malfunctioning and working at a lower range. In any case, I'm definitely not under 25 and I think I can hear at least some of these devices.
Interesting. I live near a school, and my street is a conduit for police and emergency vehicles. Once every few days I'll be sitting in my backyard and hear a sudden deafening ultra-high pitch whine for about 10 seconds, imagine tinnitus on steroids. Can't tell where it comes from. I wonder if someone's got one of these things nearby.
My friend famously said "I can see that a TV is on by just looking at it" I think he meant he could hear if the TV was on or off because he could hear the ultrasound in the electronics.
Ah the good old days. I'm not sure if I would detect a CRT today, but I remember when I was younger that I could immediately tell when there was one running anywhere nearby. You can't really hear them, but you can hear them.
what a wonderful world we're building.
Was just about to write the same. This is very much in the hearing range of most mammals
cool but also I'm a bit skeptical of ~whole body in the title. This article mentions tracking joints down to the ankles but I don't see how well it actually tracks the legs when they move independently of the upper body. What happens if you sit still and move one leg, or do a squat with your knees coming inward? Would it pick that up, or give you something closer to the squat it expects?
ive worked on a related problem, counting reps from a single chest sensor [1]. We covered 12 exercises and released the dataset on huggingface, but even avoiding false counts from movement between exercises took work. We also assumed we knew which exercise someone was doing.
anyways super curious what the acoustic signal adds when there's barely any chest or wrist movement.
[1] https://www.mdpi.com/1424-8220/24/12/3891
Yea it would be really impressive if they could do this well, I've worked with wearables as well in the past, this seems like such an interesting solution
Acoustic signal is the whole innovation here I think. the movements shown in video at top of link aren't distinguishable from accelerometer alone.
The watch is basically sonar, so distance from sensor and not (necessarily) accelerometer action is the limiting factor. Which is to say, the ankle detection could have issue, but more because it's farther from the watch than because it's not attached to your torso.
(How) would this handle multiple people using the same system being in vicinity (e.g. in the gym)? Wouldn't the speakers talk over each other?
It uses soundwaves, so while I'm no audio expert, I'd assume you could use different pitches (of inaudible sound) to listen to different users.
Different frequencies are likely going to have different signatures for each movement and require an entirely independent training set. So agree that could work, but it's probably not easy to add arbitrarily many bands for a crowded gym
Not quite to the same level, but I've recently been working on something similar with my rowing app [1].
The most common use-case is to get rowing data directly from a PM5 or FTMS rower using Bluetooth, but if you don't have a smart rower it uses Apple Watch motion to detect your strokes / stroke rate.
It detects peaks on the pull, then ignores the recovery so it doesn't detect it as a second stroke.
It adapts to how hard you're pulling as you go, so it picks up your strokes within the first few without any setup.
Overall this works well for rowing, but obviously can't accurately work out distance / power etc.
[1] Streaks Strokes: https://strokesrowing.com
Nice! I'm a Concept2 rower, what advantages does it have over the built-in software and ErgData app?
Also, you need to update your little rowing person animation at the top of the page; they're not hinging their hips enough and they're engaging their arms way too early. It's a silly little thing, but as someone who focuses on form it jumped out at me when I loaded the page.
Same thing jumped out as well as the form isn’t happening in the right order:
The Drive Phase (Way out): legs, then body hinge (back), then arms
The Recovery Phase (Way In): arms, then body hinge (back), then legs
Little rowing person is going to destroy her/his lower back given the animation
> Little rowing person is going to destroy her/his lower back given the animation
Yup
> Same thing jumped out as well
There are dozens of us! Dozens! ;-)
Thanks for the feedback - I actually traced the form for the animation from the manual (pamphlet?) that comes with the RowErg but I hear what you’re saying.
The differentiator of this app initially is more about the focus on making rowing a habit (we also make Streaks, a well-known habit tracker), and deeper integration with things like Apple Health and Apple Watch.
We’ve only just released it recently, but further customization of workouts and the display is where we’re heading.
I did something similar about ten years ago for sculling using the a mobile phone gyro to detect the scull's surge with each stroke to get a stroke rate. The code was based on an algorithm described in a paper I found online, but coded from scratch. I was rather proud of that. GPS gave an approximate speed.
Such an interesting direction, especially for older adults. I hope I live to see the day when devices like this can genuinely help our aging parents stay active and exercise safely, even in remote areas where access to a qualified trainer is limited.
Nah, they will all be frail, some earlier, some later.
It will be more the age of robots. Just look at Asia with that and you can see that they shape the future there, not just China - really, Asia. It's a bit sad to see how the USA and Europe decline in comparison here.
> Just look at Asia with that and you can see that they shape the future there, not just China - really, Asia.
Any non demo tech that's widely used there?
Not passive, noise pollution.
https://dl.acm.org/doi/epdf/10.1145/3832001
Are these loud? If so I could imagine having many people wearing these would be unbearable. I can hear very well to about 20-21kHz.
This was done with existing watch speakers and microphones. Wouldn’t ultrasound-specific speakers work better and be inaudible?
How does this work when there are multiple in the room? Does each device have its own signature so it will detect only its "waves"?
Does it work on a bed? horizontal position? This would have major functionality in hospitals.
That version would probably better be built into the bed...
would this work while swimming?
it'd be cool to track my swimming form just from the watch :)
Doubtful, at least not initially, since it uses sound waves (which travel very differently underwater).
Soundwaves travel better (less loss) underwater so if anything it would work better.
Perhaps we can now finally have a way to measure respiratory rate using a smartwatch.
I'd have imagined that light-based heart monitoring (PPG) would discern 2 fluctuating signals - pulse, and respiration.
Can't finger-tip pulse-ox monitors detect respiration derived fluctuations in oxygenation? A very brief search suggests one can.
I suspect you want a comparison of both respiratory effort and effect, however.
Respiration rate is definitely a metric that shows up in Garmin during both activities and sleep using the HR sensor via respiratory sinus arrythmia.
https://support.garmin.com/en-US/?faq=2yEgS0Pax53UDqUH7q4WC6
It gets further enhanced into breathing variations during sleep tracking if you enable the SpO2 sensor.
https://support.garmin.com/en-US/?faq=efzfZVuaKM8fLFgE36jvG9
I bet it could technically detect breathing issues like sleep apnea but they shy away from any wording that would resemble a medical diagnosis, which would immediately trigger regulatory approvals like for ECG features.
In theory you can extract respiration from RSA and possibly even by looking closely at a chest-based EEG but if you want dependable results you want a strap that can feel the pull on your chest or a radar pointed at it or something.
When you attempt biofeedback in the 1970s paradigm (look at the biosignal in real time and extend your brain’s sensorimotor maps) as compared to the 2020s paradigm (sit with a $300 device for 3 minutes, get some score and say “gee whizz!” you learn right away that many biosensors are a joke.
Ok, but can we talk about that grey body model visual...
Both unnecessarily detailed, and yet simultaneously not.
Meanwhile my smart/sportswatch sometimes thinks I'm asleep while I'm up and walking around.
next level surveillance
Have you seen the video where the inaudible passage of planes are visible in the audio spectrogram, revealing a doppler shift which can be correlated with public ADS-B databases to determine the location of the recording?
https://www.youtube.com/watch?v=UytYrhYLOKI
So its taking accelerometer data from a watch, and using it to predict body position with a model trained on whole body motion as measured by ultrasound. neat.
No, from the article: “ It uses the built-in speaker and microphone on a commercial smartwatch to determine a person’s body position using a sonar-like sensing system.” The move to commercial smartwatches uses just their single speaker and mic, rather than their custom unit they built that had dual speakers and mics.
Hmm, so does that mean if an AI LLM is trained in everything kinesiology, human body movement, anatomy, physiology, and biomechanics using this device's capabilities, it can act as an ultra physical therapist or even go as far as a physician that will actually cure their patients just by "watching" a human walk. Pun intended :)
It's quite interesting in that tiny devices are almost like ... average computers in the 1990s era. Or even better than that, now.
We could argue that the age of computing before about 2005 was the teenage years of the PC and before 2018-2020 was the teenage years for mobile devices.
Look at the rates of improvements since. For general computing they're marginal for the average person, maybe a tripling of performance every 15 years, which isn't nothing but it's also not life changing.
I.mean, follows the same principle of a Nintendo Switch/Wii controller? Games like Just Dance require certain movements and these are tracked with the Joycons
one good recommend to keep spatial position accurate an such controllers, is to keep teleporting yourself against the drift! /j
Wii required positioning with IR emitters for certain motions. Not sure about switch.
Also this is fundamentally different — those were motion sensors, this is ultrasonic position sensing — more like sonar than gyroscopes.
This intends to track the whole body by just the watch, not one 9-axis IMU each on every tracked parts.
I know this is an unpopular opinion, but this is bad for privacy especially in the hands of Big Tech.