# ELERAIQ Vision - Overview

A FastAPI-based smart camera platform. Detection and tracking run directly on the camera's onboard AI accelerator (not the host CPU), and the app serves a live dashboard over HTTP.

## Modes

| Mode | What it does |
|---|---|
| **Preview** | Clean video passthrough, no processing |
| **Object Detection** | 80-class object detector, confidence threshold + include/exclude class filters (see the Detection Classes page for the full list) |
| **People Tracking & Dwell Time** | Persistent per-person tracking (people only), dwell-time timer shown on each bounding box, zone-based "people in area" occupancy count |
| **Face Detection** | Onboard face detector |
| **Edge Detection** | Sobel edge filter, no neural net |

## Key features

- **Zones**: draggable/resizable regions drawn on the live stream; used for detection filtering, occupancy counting, and dwell-time scoping
- **Line-crossing in/out counting**: a configurable boundary line with a direction-swap toggle (for cameras mounted facing either way); daily totals persist across restarts and mode switches, auto-reset at midnight
- **Audio alerts**: trigger a sound when a chosen class is detected, "once" or "interval" cooldown modes; cross-platform sound playback on both Windows and Linux
- **Resilience**: a watchdog auto-restarts the camera pipeline if it stalls or crashes (native device/USB crashes don't raise catchable Python exceptions - detected via a stale-frame timeout instead)
- **Start/Stop control + live status** in the UI, distinct from the auto-recovery watchdog (won't fight a manual Stop)
- **Fullscreen mode** with an info overlay, collapsible sidebar, adjustable video fit (contain/cover/stretch)

## Deployment options

- **Windows (dev)**: `python app.py`, or `run_supervised.ps1` for auto-restart on crash
- **Linux, Docker**: `docker compose up --build` - see the Docker guide
- **Linux, native package**: `.deb` install with a systemd service (`Restart=always`) - see the Ubuntu Setup Guide
- **This portal**: central store for built packages, service/install files, and these docs - not a deployment target itself (no camera hardware in the cloud)

## Source

`https://github.com/yeshaib/tgcs-fr-pred`