Open-Retro-Sight

Non-invasive Industrial Edge AI Gateway - Give old machines "digital eyes"

中文版

Introduction

Open-RetroSight is a non-invasive industrial edge AI gateway software that digitizes data from traditional "dumb devices" using computer vision technology.

No disassembly, no downtime, no rewiring — Achieve digital transformation of legacy equipment with just a few hundred dollars and minutes of deployment.

Core Capabilities

• Seven-segment / LCD Display Recognition - Read temperature, pressure, counters, and other numeric displays
• Analog Gauge Reading - Map pointer angles to numerical values
• Status Indicator Light Detection - Andon light recognition for OEE calculation
• Switch / Knob Position Detection - Identify gear and toggle states

Quick Start

# Clone the project
git clone https://github.com/open-retrosight/open-retrosight.git
cd open-retrosight

# Install dependencies
pip install -r requirements.txt

# Or install via pip
pip install -e .

Documentation

For detailed documentation, see docs/:

• Introduction - Background, value proposition, use cases
• Architecture - System design and tech stack
• Use Cases - Recognition solutions for various devices
• Roadmap - Vision, MVP features, commercialization

Hardware Requirements

• Raspberry Pi Zero 2W / Orange Pi + USB Camera
• Or retired Android phone
• Or ESP32-CAM + Edge Server

Tech Stack

Python OpenCV PaddleOCR YOLOv8 MQTT Modbus TCP Streamlit

Project Status

Current Stage: MVP v0.1.0 Complete

Completed Milestones

• Product planning and design
• Documentation system established
• Project structure initialized
• Phase 1: Core capabilities (video stream, OCR, MQTT, Web UI)
• Phase 2: Key features (pointer recognition, perspective correction, Modbus, offline buffering)
• Phase 3: Extended features (indicator lights, switch recognition, image enhancement)

Roadmap

Phase	Focus	Status
v0.1	MVP Core Features	✅ Complete
v0.2	Community Building	Planned
v0.3	Hardware Kit & Product Polish	Planned
v0.4	Cloud Platform (SaaS)	Planned
v1.0	Ecosystem & Marketplace	Planned

Next Steps (v0.2)

• Build developer community
• Improve documentation and tutorials
• Collect user feedback and iterate
• Add more device recognition templates

Future Plans

• Hardware Kits: Plug-and-play kits with camera, lighting, and enclosure
• Advanced Algorithms: Complex gauge recognition, multi-target detection
• Cloud Platform: Multi-device management, data storage, alerting
• Industry Solutions: Manufacturing, energy, chemical, logistics

Phase 1 Implemented Features

• Video Stream Capture (src/retrosight/capture/camera.py)

  • USB/CSI/RTSP camera support
  • Multi-threaded frame buffering

• Digital OCR Recognition (src/retrosight/recognition/ocr.py)

  • Seven-segment display recognition (PaddleOCR)
  • Image preprocessing optimization

• Data Smoothing Filters (src/retrosight/preprocessing/filter.py)

  • Kalman filter, moving average, exponential smoothing
  • Outlier filtering

• MQTT Data Publishing (src/retrosight/output/mqtt.py)

  • Auto-reconnect, offline caching
  • Structured JSON data

• Web Configuration Interface (src/retrosight/ui/app.py)

  • Streamlit real-time preview
  • ROI region configuration

Phase 2 Implemented Features

• Perspective Transform Correction (src/retrosight/preprocessing/transform.py)

  • Four-point perspective transform
  • Image registration (vibration resistant)
  • Lens distortion correction

• Pointer Recognition Algorithm (src/retrosight/recognition/pointer.py)

  • Hough line transform for pointer detection
  • Automatic dial center detection
  • Angle to value mapping

• Modbus TCP Output (src/retrosight/output/modbus.py)

  • Modbus TCP Server (PLC emulation)
  • Multiple data type support (INT16/FLOAT32, etc.)
  • Automatic register allocation

• Offline Buffer & Retry (src/retrosight/output/buffer.py)

  • SQLite persistent storage
  • Priority queue
  • Auto-retry and expiration cleanup

Phase 3 Implemented Features

• Indicator Light Recognition (src/retrosight/recognition/light.py)

  • HSV color detection (red/yellow/green/blue/white)
  • Andon tower light status monitoring
  • Blink detection and frequency estimation
  • OEE availability calculation

• Switch / Knob Recognition (src/retrosight/recognition/switch.py)

  • Toggle switch ON/OFF detection
  • Rotary knob position recognition (multi-position support)
  • Push button / slide switch states
  • Template matching and position detection

• Image Enhancement (src/retrosight/preprocessing/enhancement.py)

  • Adaptive contrast enhancement (CLAHE)
  • Glare removal (inpainting)
  • Multi-frame fusion denoising
  • Low-light enhancement

License

MIT License

Related Links

• Documentation Center
• Product Vision
• Technical Challenges