🧠 THYX: AIoT Edge Intelligence Video Analysis Platform

THYX is an edge-side intelligent video analysis platform designed for AIoT scenarios. It integrates video stream access, behavior recognition, intelligent alerting, and visual management, covering the full workflow from algorithm configuration to concurrent multi-task analysis. The platform features high-performance inference, a modular architecture, and supports lightweight deployment across various edge devices with flexible scene combinations and hot-pluggable extensions.

With built-in voice announcements, pop-up alerts, real-time video display, and a web management backend, THYX delivers an intuitive and efficient human-computer interaction experience. It is widely applied in industries such as security, manufacturing, and transportation, helping build stable and reliable intelligent perception systems.

---

Demo | Chinese Document | Video Introduction

---

📸 Platform Features Overview

THYX provides a full end-to-end workflow for intelligent video analysis, from video stream access to intelligent alerts and visual presentation. Key features include:

• Flexible dynamic license management for platform-level and scenario-level algorithms
• Supports RTSP, RTMP, and local video access with unified multi-stream management
• Concurrent multi-task processing with customizable real-time analysis per stream
• Real-time alerts (voice + pop-up) and searchable historical alert records
• Real-time display of scene analysis per stream for intuitive understanding
• Visual network configuration interface for easier deployment
• Online dynamic adjustment of algorithm thresholds, with quick access to algorithm list and descriptions
• Real-time configurable global parameters

✅ Sample Scenario Capabilities (more available as plug-ins)

Scenario Name	Application Scene	Detection Example Image
Helmet Detection	Construction sites, factory safety
Fire Detection	Forest fire prevention, factories
Smoking Detection	Gas stations, hospitals, plants
Fighting Detection	Campuses, malls, subways, prisons
Post Detection	Duty rooms, production lines

Dozens of scenario algorithms are already implemented. See: Algorithm Config / Algorithm List

---

🎯 Key Advantages

• High-Performance Edge Inference
  Concurrent multi-stream processing with <200ms per-stream latency; supports 1080P real-time analysis
  

• Modular Architecture
  Plugin-based modules support hot-plugging and hot updates. Unified AI inference service with flexible alarm management
  

• Full-Scenario Coverage
  Covers security, traffic, industrial, fire protection, etc. Includes plug-and-play pre-trained models and supports rapid integration of custom scenes
  

• Smart Interaction
  Provides a visual task configuration UI, video display, real-time voice alerts and pop-up notifications
  

---

🚀 Quick Deployment Guide

1. Clone the Repository

sudo mkdir /data && cd /data
git clone https://github.com/senthree3/THYX.git
cd THYX

2. Install the Python Virtual Environment

Download the pre-built environment venv.tar.gz and extract it into the project directory:

tar -xzf venv.tar.gz

A venv/ folder should appear after extraction

3. Download Model Weights

Download the model package, extract it into the project root, and rename to plugins/:

unzip models.zip -d ./plugins

4. Run Deployment Script

chmod +x deploy.sh
sudo ./deploy.sh

5. Collect Static Files

source /opt/ai_box/venv/bin/activate 
python manage.py collectstatic

6. Reboot Device

sudo reboot

7. Access the Platform

Open your browser and visit:

http://<device-ip-address>

Default credentials:

• Username: admin
• Password: admin

The platform itself is hardware-agnostic. The provided algorithm plugins currently support SOPHGO BM1684 series. For other chipsets, cross-compilation and model adaptation is required.

---

🧩 Core Modules

📷 Channel Management

Supports RTSP, RTMP, local video file access with unified stream management

📊 Task Center

Configure multiple scene algorithms per stream. Supports dynamic parameter tuning, 3rd-party alerting, region setup, and task monitoring

🚨 Alarm Center

Multi-dimensional search (by channel, time, type), with event details and playback

🖥️ Real-Time Screen

Displays real-time AI analysis of all streams for intuitive situational awareness

---

🛠 Tech Stack Overview

• Deployment: Bare-metal edge deployment, no cloud dependency

---

📦 Project Structure

THYX/
├── ai_box/               # Middleware and config
├── bm_ai/                # Algorithm scheduling core
├── content/              # Runtime cache
├── logs/                 # Log files
├── media/                # Media assets
├── mosquitto-offline/    # Offline MQTT data
├── plugins/              # Algorithm plugins (hot-pluggable)
├── server/               # Nginx config
├── static/               # Static files (JS, CSS, images)
├── templates/            # HTML templates
├── venv/                 # Python virtual environment
├── ai_box.db             # SQLite database
├── deploy.sh             # One-click deployment script
├── manage.py             # Django main script
├── restart_all.sh        # Restart all services
├── run_all.sh            # Start all services
├── stop_all.sh           # Stop all services
├── README.md             # English README
└── README_ZH.md          # Chinese README

---

📝 License

This project is licensed under the Apache License 2.0. See LICENSE for full details.

---

🙌 Acknowledgments

This platform is based on the sophon-stream video engine and built with Django.
Special thanks to the community and partners for their continuous support of AIoT video intelligence applications.

---

📮 Contact

• Project: https://github.com/senthree3/THYX
• Issues: GitHub Issues
• Email:
  • senthree30@gmail.com
  • 717192305@qq.com