Comment by transpute
1 year ago
Through-wall 2.4Ghz Wi-Fi CSI radar can be done with $20 ESP32 boards, https://www.technologyreview.com/2024/02/27/1088154/wifi-sen...
1 year ago
Through-wall 2.4Ghz Wi-Fi CSI radar can be done with $20 ESP32 boards, https://www.technologyreview.com/2024/02/27/1088154/wifi-sen...
Wifi sensing is interesting, but you can combine a $3 ESP chip with one of these for a much easier project at lower cost: https://www.dfrobot.com/product-2795.html
I use those for presence detection in my house. 3D print a small case and for < $20 you have highly accurate presence detection that also works when sitting still. Ideal for automatic light and climate control.
My problem with similar sensors is that they don't work very well if you're sitting still around 3m away. Maybe my sensor isn't great, but when I'm sitting still at the computer, it frequently thinks I'm not there any more, which ruins any benefit.
There are some that detect heartbeat, I guess those might work. But those are mmWave, not just cheap 2.4GHz radars or PIR.
4 replies →
A hack is to put a sensor on the chair
8 replies →
Pairing them with PIR detection should help to also detect stationary warm objects.
2 replies →
The “Vital Sign Detection & Healthcare”[1] section is very interesting, thank you! I am looking for a way to measure the heartbeat of a single person in a completely non-invasive manner (so without contact) and I see that it can be done with Wi-Fi (2019 Paper [2]). However, I’ve noticed that there are a lot of methods (video analysis, thermal camera, etc.), but they are hard to find as ready-made products at a reasonable price. The simplest solution is under-mattress sleep trackers, but unfortunately they are not an option for me.
1. https://github.com/Marsrocky/Awesome-WiFi-CSI-Sensing#vital-...
2. https://arxiv.org/abs/1908.05108
> measure the heartbeat of a single person in a completely non-invasive manner
NIST, "Monitoring Respiratory Motion with Wi-Fi CSI" (2022), https://www.nist.gov/news-events/news/2022/12/wi-fi-could-he... & https://doi.org/10.1109/ACCESS.2022.3230003
Google: Nest Hub sleep sensing for one person, with Fitbit Premium, https://www.techradar.com/health-fitness/how-to-track-your-s...
Interesting. I'd be interested in a breathing sensor that can work on someone working at a computer.
I’m sure it can be done with a normal camera and enough light, especially with the advancement of AI technologies in recent years. I haven’t had time to experiment, but if you have an iPhone, you can try this app [1] I installed yesterday to study its feasibility.
On the homepage, there is a video [2] explaining more as well, and their paper [3].
1. https://www.rouast.com/vitallens/
2. https://www.youtube.com/watch?v=_0HHjovI8hc
3. https://arxiv.org/pdf/2312.06892
Intel Meteor Lake+ platforms can use NPU for WiFi 7 Sensing.
Presence (commercial): https://community.intel.com/t5/Blogs/Tech-Innovation/Client/...
Breathing (research): https://www.intel.com/content/www/us/en/research/respiration...
For older Intel wifi radios, there is custom firmware and many research papers: https://dhalperi.github.io/linux-80211n-csitool/
> chewing of people or animals
The ESP32 can detect the difference between a cannibal and a vegan?
That's impressive.
How can you tell someone is a vegan? Don't worry, this 1$ radar motion sensor can tell you.
I think you could go cheaper. A sound detection sensor is a bit cheaper, and it just needs to detect the word 'vegan.'
11 replies →
Come on, that's simple, just use voice recognition since the vegan won't stop telling the sensor about it.
Great links! I recall seeing somewhere that AMEX was investing in WiFi object detection…can’t remember exactly for what or how…
You can also use a single ESP and your WiFi AP as a source of packets.
These $1 boards also work through walls (I tried this).
I always go through walls.
— kool aid man
The gesture recognition sounds incredibly interesting. I work with vision based band tracking, and performance really isn't good enough for my application.
I wonder what kind of resolution you can reasonably achieve? Is it good enough to detect finger pose?
Aside: is there a way around IEEE paywall? I'd really love to read some of these papers