CV Safety System

A computer-vision safety monitoring system for exhibition environments, combining relic detection/tracking, human pose estimation, dangerous-object detection, and real-time alert visualization in a single video pipeline.

The current default configuration treats cup as the protected relic class and knife, scissors, and baseball bat as dangerous classes (configurable in code).

Features

• YOLOv7-tiny detection with lightweight multi-object tracking and relic selection logic
• MediaPipe Pose 33-keypoint estimation
• Relic fence intrusion detection plus dangerous-object/person association alerts
• PySide6 desktop UI with alert list, status panel, and video overlays
• Automatic first-run download for pose model and YOLO weights

Requirements

• Python 3.10+
• Linux / macOS / Windows (with camera access permissions)
• Internet access for first-time model download

Installation

# 1) Install dependencies
pip install -r requirements.txt

# 2) Clone YOLOv7 repository
# The current implementation dynamically imports inference utilities from this local folder.
git clone --depth 1 https://github.com/WongKinYiu/yolov7.git

Quick Start

# Recommended: launch the integrated PySide6 client
python run.py --source 0

Optional arguments:

• --source: camera index (for example, 0) or video file path
• --conf: YOLO confidence threshold (default: 0.25)
• --pose-model: pose model path (default: models/pose_landmarker_full.task)
• --yolo-model: YOLO weights path (default: models/yolov7-tiny.pt)
• --alert-sound: custom alert sound file path

Alternative Entry Points

# Integrated monitoring logic with OpenCV window
PYTHONPATH=src python -m cv_safety_sys.monitoring.integrated_monitor --source 0

# Directly launch the Qt module
PYTHONPATH=src python -m cv_safety_sys.ui.qt_monitor --source 0

# Detector/tracker only (debug)
PYTHONPATH=src python -m cv_safety_sys.detection.yolov7_tracker --source 0

Repository Structure

cv_safety_sys/
├── run.py
├── requirements.txt
├── src/cv_safety_sys/
│   ├── detection/           # YOLOv7 detection and tracking
│   ├── monitoring/          # Relic + pose + dangerous-object safety logic
│   ├── pose/                # MediaPipe pose model download helper
│   ├── ui/                  # PySide6 client
│   └── utils/               # Utilities (for example, text rendering)
└── docs/                    # Architecture and module documentation

Documentation

• System architecture: docs/system_architecture.md
• Relic protection workflow: docs/object_protection.md
• Pose module guide: docs/webcam_pose_detection.md

License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0). You may copy, modify, and redistribute this project under the terms of GPL-3.0. See the LICENSE file for details.

Open Source Notice

This project is released under GPL-3.0. You may use, modify, and redistribute it under GPL-3.0 conditions. If you distribute modified versions, you are generally required to provide corresponding source code under GPL-compatible terms. See LICENSE for full terms.

Copyright

Copyright (C) 2025 Linsheng Yin, Heng Quan, Bojin Li, Yunpeng Din, Penghan Chen