Starting in the early aughts and extending into the release of the Wiimote there were a handful of projects like this, largely for Mac. Never amounted to much more than a proof of concept most of the time but I always found the concept exciting. I had a lot of hope for Magic Leaps initial stand-alone hand tracking device before they pivoted to AR
Are you by chance interested in a command named after the four-letter word, which, when invoked, automatically fixes typos in the last run command and reruns it: https://github.com/nvbn/thefuck
In the screenshot they’ve mapped fist to “shutdown now” which seems to either indicate high confidence in the gesture recognition capabilities, or a good sense of humor about the whole thing.
I worked on something very similar to this a couple years ago. My main issue with it why I didn't go beyond the PoC stage was that raising my hands up to the camera soon became a pain (I use a laptop), esp as I have it at eye level. It also had a hard time keeping up with my motions, so most of the time it was getting a blur.
I had a overhead-ish camera on a stick, that flopped up from the laptop screen back, to point straight down at the keyboard. Alternately, a phone clip-on fish-eye lens was low-res but ok, iirc. Kludgy little mirrors were iffyer, but there's a commercial one for iPad I didn't try.
Blur wise, perhaps it was 30 fps? I was mostly ok with 60 fps. Would have liked 4k resolution for markers and precision though.
More problematic was integrating high-latency "wait for the camera frame and then for mediapipe" with low-latency "fast keypress response". Drove me towards replayable event streams, and thus custom apps. But I was using it for custom XR-ish apps anyway.
Running high-res mediapipe etal on multiple cameras kept a thinkpad doorstop laptop warm even before starting XR. Keyboard feel got a bit toasted. But TFA doing gestures-only fixed-poses low-res single-camera is far lighter.
It's a pretty old system; had it over 10 years now, so I didn't expect too much re the blur. I'll likely revisit after I've upgraded to something more modern. Maybe by then I'll have the camera position figured as well.
20 comments
[ 5.9 ms ] story [ 59.7 ms ] threadThe rest is just some glue.
I feel like a few gestures to lock or switch workspaces would be really ideal.
Blur wise, perhaps it was 30 fps? I was mostly ok with 60 fps. Would have liked 4k resolution for markers and precision though.
More problematic was integrating high-latency "wait for the camera frame and then for mediapipe" with low-latency "fast keypress response". Drove me towards replayable event streams, and thus custom apps. But I was using it for custom XR-ish apps anyway.
Running high-res mediapipe etal on multiple cameras kept a thinkpad doorstop laptop warm even before starting XR. Keyboard feel got a bit toasted. But TFA doing gestures-only fixed-poses low-res single-camera is far lighter.
Love to get into the XR space myself.
1: https://en.wikipedia.org/wiki/Finger_gun
https://github.com/bjesus/eyeroll/