👥 Crowd-Density-Detection

This project is based on a Deep learning-based crowd density estimation system using CrowdNet CNN architecture to analyze crowd images, generate density maps via Gaussian filter transformation, and classify public spaces as safe or unsafe based on predicted crowd counts with safety threshold categorization.

Key Features

• Implements CrowdNet CNN architecture for spatial feature extraction from crowd images.
• Analyzes two image datasets for crowd density estimation: ShanghaiTech Dataset (1198 images) and the Mall Dataset (2000 images)
• Generates density maps using Gaussian Filter Transformation.
• Safe vs. Unsafe crowd levels based on threshold analysis

Tech Stack

• Python - Primary programming language
• TensorFlow/Keras - Deep learning framework for CNN implementation
• OpenCV - Image Processing and Transformation
• Pandas - Data manipulation and preprocessing
• NumPy - Numerical computing and array operations
• Matplotlib / Seaborn - Data visualization
• scikit-image - Gaussian filter transformation for density maps

How it Works?

Data Collection

• Utilized the ShangahiTech Image Dataset and the Mall Image Dataset containing crowd scenes.
• Images captured different crowd densities, lighting conditions, and environments.
• The dataset includes both safe and unsafe crowd density scenarios.
  

📁GET THE DATA

ShanghaiTech [Part A and Part B] Dataset

Mall Dataset

Data Preprocessing

• Resized and normalized crowd images for consistent input dimensions.
• Applied Gaussian Filter Transformation to generate density maps from crowd images.

Model Comparison and Training

• Implemented CrowdNet CNN architecture designed for crowd analysis.
• Pooling layers reduce dimensionality while preserving critical spatial information.
  

Model Architecture

Crowd Density Prediction

• Model processes input crowd images through CrowdNet architecture
• Generates predicted crowd counts and density estimates
• Density maps visualize spatial distribution of people in the scene
• Outputs numerical crowd count predictions for each image

Safety Classification

• Established safety threshold based on crowd density analysis.

Performance Evaluation

• Tested model on separate test dataset to evaluate generalization.

Visualization

• Scatter plots: Actual vs. Predicted crowd counts showing model accuracy.
• Density map visualizations highlighting crowd concentration areas.

Design Framework

Screenshots

Density Map Generation:

Model Training:

Model Testing:

Safety Threshold-based Categorization:

Comparative Analysis of Crowd Density:

PART A:

PART B:

Note

Data Access: Due to licensing and privacy considerations, datasets are not included in this repository