Simple JStures can go a long way
summary
Live gesture recognition works end-to-end: an Arduino Nano 33 BLE Sense streams accelerometer and gyroscope arrays to the browser, a tiny TensorFlow.js model (~8 KB) trains and exports via tiny motion trainer, and the browser transfers the model over Bluetooth to run inferences on the board. Claims: prototype runs in minutes, fits in under ~50 lines of JS (sometimes <15), transfers model in chunks with checksums, and demos (air Street Fighter, air drums) actually work.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 2:17Human computer interaction
- 3:44Motion detection possibilities
- 7:02Hardware requirements
- 8:40Training gesture models
- 14:10Integrating into web apps
- 20:36Creative demos and applications
related talks
28mStrike a Pose - Gesture recognition in JavaScript w/ ML & Arduino
22mStrike a Pose: Gesture Recognition in JS with Machine Learning
55mBuilding with TensorFlow Lite for microcontrollers
37mJavaScript On Microcontrollers
54mSmarter Everything with ML & the IoT
32mWe Also Can Do It! Machine Learning in Javascript!
32mDeep Learning in JS
39mMachine Learning Magic for Your JavaScript Application