Artificial Intelligence in Small Embedded Systems
summary
Describes moving AI inference to edge devices to avoid raw video transfer by shipping compact trained models instead. Presents a prototype camera+mic+battery+display running ~8 MB quantized face and action recognition models on low‑cost accelerators (Coral USB ~ $35, older Intel sticks ~ $100), achieving ~40–60 FPS per stick and sometimes using four sticks in parallel. Demonstrations include local door unlocking, fall/pool‑child alerts, and operation without uploading video.
AI-written summary · about
chapters (11)
- 0:00AI for embedded intro
- 1:46Image recognition history
- 3:31AI trends in embedded
- 5:18Intel Movidius Stick apps
- 11:30AI and IoT synergy
- 13:40AIoT architecture examples
- 17:13Fog computing for smart cities
- 19:44AIoT configurator
- 23:38Universal device demo
- 29:52Q&A
- 38:44Summary
