Factories and farms have sensors everywhere — IP cameras, MQTT sensors, PLCs, audio microphones, cellular cameras — but they're all siloed on different networks with different protocols. Alert fatigue is rampant, financial systems are disconnected, and there's no single view of what's actually happening.
BatchDrone is a 5-layer distributed platform that unifies any sensor into a single intelligent dashboard. Each edge device learns what "normal" looks like for its zone (zero calibration) and alerts on anomalies using statistical Z-score detection. For farms, it adds AI livestock recognition with per-animal economics. For factories, it connects sensor data directly to ERP and accounting systems.
Layer 5: INTEGRATION ─────── Financial Systems (Sage, Xero, SAP, Syspro)
│
Layer 4: INTELLIGENCE ────── Tracking, Anomaly Rules, AI Recognition
│
Layer 3: AGGREGATOR ──────── Central Cloud Hub (FastAPI + SQLite)
│
Layer 2: EDGE DEVICES ────── On-Site Processing (Raspberry Pi, Intel NUC)
│
Layer 1: SENSORS ─────────── IP Cameras, MQTT, HTTP, GPIO, Audio, WebRTC
Every camera zone is continuously analyzed across 6 dimensions — no calibration required:
| # | Metric | Detects |
|---|---|---|
| 1 | Luminance Mean | Lights on/off, flashlights, darkness |
| 2 | Color Entropy | Smoke, fire, chaotic movement |
| 3 | Edge Density | Camera focus loss, lens obstruction |
| 4 | Texture Uniformity | Spills, defects, breakage |
| 5 | Grid Hotspots | Localized intrusion or movement |
| 6 | Complexity Score | Sudden appearance of detailed objects |
Anomaly detection: Z-score analysis (>3σ = alert). Each zone self-learns its baseline over 24 hours.
Smart frame sampling: ~90% CPU savings by skipping redundant frames.
Raw anomalies are transformed into plain-English events:
| Signal Combination | Interpretation |
|---|---|
| Luminance drop + complexity spike | Intrusion detected |
| Color entropy spike + texture change | Fire/smoke detected |
| Edge density collapse | Camera obstruction |
| Grid hotspot + complexity | Unauthorized entry in Zone 3 |
Configurable signatures via API. Automatic action dispatch (IoT alarms, API calls, notifications).
- Interactive floor plan with drag-drop asset placement
- Multi-sensor aggregation per asset (camera + MQTT + PLC + audio)
- Event rules engine with real-time state colors
- Per-product cost tracking linked to ERP
- Leaflet GPS satellite map with camera zones and animal markers
- AI animal recognition (species, breed, sex, color via Grok/OpenAI vision)
- 128-dimensional facial embedding for individual animal identification
- Herd management with sighting feeds and missing-animal alerts
- Per-animal P&L tracking
| Layer | Technology |
|---|---|
| Cloud | FastAPI on GCP Compute Engine |
| Edge | Flask + OpenCV (Raspberry Pi / Intel NUC) |
| Database | SQLite + SQLAlchemy Async (17+ models) |
| Mapping | Leaflet.js with satellite imagery |
| Vision AI | Grok/OpenAI for livestock recognition |
| Deployment | systemd services, JSON queue persistence |
| Protocols | ONVIF, RTSP, MQTT, HTTP, GPIO, WebRTC |
- Factories — Manufacturing plants with sensor blind spots and alert fatigue
- Farms — Automated headcounts, individual animal tracking, per-animal economics
- Warehouses — Perimeter security with anomaly detection
- Multi-site operations — Connecting disparate sensors across locations
| Feature | BatchDrone | Traditional CCTV | Generic IoT |
|---|---|---|---|
| Self-learning | Zero calibration | Manual rule setup | Threshold config |
| Sensor-agnostic | Any protocol, any device | Camera-only | Platform-locked |
| Offline edge | Works without internet | Cloud-dependent | Cloud-dependent |
| Financial integration | Sensor → ERP/Accounting | None | None |
| Explainable alerts | Z-scores + plain English | Motion rectangles | Raw values |
Visit batchdrone.com to explore the platform and request a demo.
Built by F² AI . Deployed globally on Google Cloud Platform.
MIT