pi5-hailo-vision

Self-hosted multi-camera AI vision for Raspberry Pi 5 (16GB) + Hailo-10H AI HAT.

• YOLO11s object detection on Hailo (real-time, multi-stream)
• CLIP text/image embeddings for "find me clips of X" semantic search
• Multi-object tracking (IDs persist across frames)
• RTSP ingest for any WiFi/PoE IP camera
• FastAPI + WebSocket backend, modern web UI
• Tailscale for secure remote access from anywhere
• systemd service, auto-starts on boot

---

Quick start

1. Flash SD card → see docs/01-flash-and-boot.md
2. Install everything:

   git clone https://github.com/rgthelen/newtestapp.git ~/pi5-hailo-vision
   cd ~/pi5-hailo-vision
   ./scripts/install-all.sh

3. Add cameras → edit ~/.config/pi5-hailo-vision/cameras.yaml
4. Start service:

   sudo systemctl enable --now pi5-hailo-vision

5. Open the UI at http://<pi-tailscale-name>:8080

---

What's in this repo

scripts/         # one-shot installers (OS, Hailo, models, Tailscale)
app/             # Python backend (pipeline + API + UI assets)
  pipeline/      # RTSP → Hailo YOLO → tracker → CLIP embeddings
  api/           # FastAPI routes + WebSocket live feed
  storage/       # SQLite + snapshot store + embedding index
  ui/            # static web UI (vanilla JS, no build step)
config/          # example camera + model config
systemd/         # service unit
docs/            # setup walkthroughs

---

Docs

• 01 — Flash the SD card and first boot
• 02 — Install Hailo stack
• 03 — Models (YOLO11s + CLIP HEFs)
• 04 — Tailscale
• 05 — Add cameras (RTSP)
• 06 — Web UI
• 07 — Troubleshooting