Crowd Detection Using CV

Overview:

In public spaces, the inability to perceive crowd density poses a significant risk to individuals with visual impairments. This project addresses that challenge by developing a real-time crowd detection system, enabling safer navigation through the environment.

Objective:

• To design a reliable and efficient system that detects crowded versus non-crowded environments from images and provides actionable alerts to assist visually impaired users in making safer mobility decisions.

Key Features:

• Real-Time Crowd Detection:

  • Utilizes computer vision techniques to classify crowd levels in real-time.

• High Accuracy Classification:

  • Achieved highest accuracy using SVM, outperforming classifiers like KNN, Random Forest, Decision Tree, and Logistic Regression.

• Assistive Integration:

  • Designed for future integration with wearable devices to provide audio or haptic feedback to the user.

• Multi-Classifier Evaluation:

  • Benchmarked multiple ML models to ensure optimal performance across different scenarios.

Methodology:

• Data Collection & Preprocessing:

  • Compiled a dataset of images labeled as crowded or non-crowded.

  • Applied preprocessing steps including resizing, grayscale conversion, and histogram equalization.

• Feature Extraction:

  • Used SIFT (Scale-Invariant Feature Transform) to extract robust visual features.

  • Reduced feature dimensions with Principal Component Analysis (PCA) to enhance model efficiency.

• Model Training & Evaluation:

  • Trained multiple classifiers including SVM, Random Forest, KNN, Decision Tree, and Logistic Regression.

  • Evaluated models using accuracy, precision, recall, and F1-score.

Results:

• The system achieved high accuracy with SVM for crowd classification.

• Demonstrated potential for real-time crowd awareness assistance for visually impaired individuals.

Future Work:

• Dataset Expansion: Incorporate varied crowd types, perspectives, and lighting conditions.

• Real-World Testing: Conduct user trials to validate system usability in live environments.

Link to the dataset : https://drive.google.com/drive/folders/1-rUfOo3d-FGsb7si3mUwA_prOiAK4usc?usp=sharing