BRT Crowd Analysis System

📋 Overview

An advanced crowd analysis system for Peshawar's Bus Rapid Transit (BRT) using YOLOv8 and computer vision techniques. The system performs real-time crowd detection and monitoring to improve safety and efficiency in BRT stations.

• Accuracy: 98% in detecting and counting people within specified regions of interest
• Purpose: Enables proactive crowd management by providing accurate monitoring data
• Application: Optimizes resource allocation and prevents overcrowding situations

Note: The BRT video footage used in this project is strictly for educational and research purposes only. The footage should not be shared online or used in any misleading way. All rights to the footage belong to their respective owners.

✨ Key Features

• Real-time Detection & Counting

  • Accurate people detection using YOLOv8
  • Custom region of interest (ROI) definition
  • Threshold-based crowd monitoring

• Advanced Visualization Interface

  • Modern UI with professional header/footer
  • Color-coded status indicators (Normal/Warning/Critical)
  • Comprehensive statistics panel
  • Real-time count history graph
  • Semi-transparent ROI visualization
  • Elegant detection boxes with corner highlights

• Developer Tools

  • Frame extraction utility for dataset creation
  • Support for custom trained models
  • Detailed performance metrics

🖼️ Visualization Interface

The system features a modern and informative visualization interface designed for effective crowd monitoring:

Display Elements

Component	Description
Header Bar	Professional title display with clean, modern design
Counting Information	Real-time count with color-coded progress bars
Statistics Panel	Shows current/max/avg counts, FPS, and runtime
Count History Graph	Real-time mini-graph showing count trends
Detection Visualization	Elegant bounding boxes with corner highlights
Footer Information	Timestamp and copyright information

Color Scheme

Status	Color Code	Usage
Normal	Green (#00FF00)	Below 60% of threshold
Warning	Orange (#FFA500)	Between 60-90% of threshold
Critical	Red (#FF0000)	Above 90% of threshold
Accent	Gold (#FFCC00)	UI highlights and borders

🚀 Getting Started

Prerequisites

• Python 3.7+
• CUDA-compatible GPU (recommended for real-time processing)

Installation

1. Clone the repository

   git clone https://github.com/hassanrrraza/crowd-analysis-yolo.git
   cd crowd-analysis-yolo

2. Install dependencies

   pip install -r requirements.txt

3. Download model weights (if not included)

   • Place YOLOv8 weights files in the models/ directory
   • Or use the included best.pt custom trained model

💻 Usage

Command Line Interface

# Basic usage
python main.py

# With custom parameters
python main.py --model best.pt --video cr.mp4 --threshold 50

# Show all available options
python main.py --help

Python API

from src.people_counter import PeopleCounter

# Initialize the counter
counter = PeopleCounter(
    model_path="models/best.pt",
    video_path="data/videos/cr.mp4",
    class_file="data/coco1.txt",
    threshold=40
)

# Run with visualization
counter.run(display=True)

Extracting Frames

python src/utils/image_extractor.py --video data/videos/cr.mp4 --output data/images/extracted --max-frames 200

📁 Project Structure

crowd-analysis-yolo/
├── data/                  # Data files
│   ├── videos/            # Video files for analysis
│   ├── images/            # Extracted frames and training images
│   ├── labels/            # Annotation files for training
│   └── coco1.txt          # Class names file
├── models/                # Trained YOLO model files
├── notebooks/             # Jupyter notebooks for experimentation
├── src/                   # Source code
│   ├── people_counter.py  # Main people counting module
│   ├── utils/             # Utility functions
│   │   └── image_extractor.py  # Frame extraction utility
│   └── models/            # Model loading and management
├── requirements.txt       # Project dependencies
└── README.md              # Project documentation

🔧 Training Custom Models

For custom model training, refer to the Jupyter notebook notebooks/yolov8_object_detection_on_custom_dataset.ipynb which provides step-by-step instructions for:

1. Preparing your dataset
2. Configuring training parameters
3. Training the model
4. Evaluating performance
5. Exporting for inference

🙏 Acknowledgments

• Ultralytics YOLOv8
• CVZone
• OpenCV

👤 Author

Hassan Raza

• Email: [email protected]
• LinkedIn: hassanrrraza
• GitHub: hassanrrraza

📄 License

© 2025 Hassan Raza. All rights reserved. This project is available for use under the MIT license. You are free to use, modify, and distribute this code in your work, provided that you give appropriate credit to the original author.

If you use this project in your work, please cite it as: "BRT Crowd Analysis System using YOLOv8 by Hassan Raza".