Artificial Vision Project 2024/2025

Project Description

This project, part of the Artificial Vision Contest 2025, aims to develop an advanced system for automatic video analysis. The proposed software will: Detect and track all people in real-time within a video. Recognize pedestrian attributes such as gender, bags, hats. Analyze behaviors by monitoring virtual line crossings.

Refer to Report.pdf for the detailed explaination and analysis of the system.

Goals

Ensure high precision in detection and classification to minimize errors. The system will process videos in real-time, display results through an interactive interface, and generate a standardized output file.

Project Structure

├── classification/                                # Classification files
│   ├── dataset.py                                 # Dataset handling for classification tasks
│   ├── nets.py                                    # Model definitions and architectures for pedestrian attribute classification
│   ├── preprocess.py                              # Data preprocessing pipeline
│   ├── train_strategy1.py                         # Training script for strategy 1
│   ├── train_strategy2.py                         # Training script for strategy 2
│   └── test.py                                    # Testing and evaluation script
│
├── confs/                                         # Configuration files
│   ├── botsort.yaml                               # Tracker configuration for BoT-SORT
│   └── config.txt                                 # Configuration files for camera and lines
│
├── dataset/                                       # Dataset-related files
│
├── models/                                        # Models directory
│   ├── yolo11m.pt                                 # Pre-trained YOLO model for pedestrian detection
│   ├── classification_model_strategy1.pth         # Trained model for classification with strategy 1
│   └── classification_model_strategy2.pth         # Trained model for classification with strategy 2
│
├── result/                                        # Result files
│   └── result.txt                                 # Result file to store analyzed results
│
├── videos/                                        # Video files for testing and analysis
│
├── gui_utils.py                                   # GUI utilities for rendering bounding boxes, text, and other visual elements
├── lines_utils.py                                 # Utilities for managing and checking line crossings
├── main.py                                        # Main script for running the pedestrian analysis system
├── OutputWriter.py                                # Manages saving results and generating reports
├── tracking.py                                    # Tracking-related functions using BoT-SORT
└── README.md                                      # Project documentation

Requirements

Programming Language : Python 3.+

Dependencies

torch torchvision ultralytics opencv-python Pillow numpy matplotlib scikit-learn tqdm seaborn

Recommended Hardware

GPU with CUDA support for optimal performance.

Execution

- To run the project on the example video "videos/Atrio.mp4", execute the main file: python main.py

• Full datasets are avaiable at: Training images https://drive.google.com/file/d/1uEcO7zgZilzDhbr1wdGkoG6LxH_uJ8ay/view?usp=share_link Validation images https://drive.google.com/file/d/1HXJdXgnjYb2AcHO841McnUlw4roJthK-/view?usp=share_link

• Trained models are not avaiable