Human Anomaly Detection Using Multilevel Classification using UCF Crime Dataset with 3 different models(CNN, VGG16, AutoEncoders)

Project Overview

This project focuses on Human Anomaly Detection in video surveillance footage using advanced machine learning and deep learning techniques. Leveraging the UCF Crime Dataset, which comprises 14 distinct crime classes, the system is designed to detect and classify human anomalies in real-world scenarios effectively. The project's primary goal is to improve the accuracy and reliability of surveillance systems for detecting criminal activities.

Key Features

• Dataset Used: UCF Crime Dataset, containing real-world surveillance footage categorized into 14 crime types.
• Multilevel Classification: A hierarchical approach for anomaly detection and classification to improve precision and accuracy.
• Deep Learning Models:
  • Convolutional Neural Networks (CNNs): Applied for feature extraction and classification.
  • VGG16: A pre-trained deep learning model fine-tuned for anomaly detection tasks.
  • Autoencoders: Used for unsupervised anomaly detection by reconstructing frames and identifying deviations.
• High Accuracy: Achieved an overall accuracy exceeding 99% in classifying anomalies across the 14 crime categories.

Methodology

1. Data Preprocessing:

   • Extracted relevant frames from surveillance videos.
   • Performed data augmentation to enhance model robustness.
   • Normalized input data for efficient model training.

2. Model Architecture:

   • CNNs were trained to identify spatial patterns and features in video frames.
   • VGG16 was fine-tuned to classify anomalies with higher accuracy by leveraging transfer learning.
   • Autoencoders were employed to detect anomalies by identifying significant reconstruction errors.

3. Training and Validation:

   • Split dataset into training, validation, and test sets in 80,20,20.
   • Used cross-validation techniques to prevent overfitting.
   • Monitored training performance using accuracy, precision, recall, and F1-score.

4. Evaluation:

   • Measured model performance using metrics like accuracy, confusion matrix, and ROC-AUC curves.
   • Compared results across different models to determine the most effective approach.

Results

• Overall Accuracy: Achieved over 99% accuracy in detecting and classifying human anomalies across 14 crime categories.
• Model Comparison:
  • CNN: High performance in basic anomaly detection tasks.
  • VGG16: Superior classification accuracy for complex scenarios.
  • Autoencoders: Excellent at identifying subtle anomalies through unsupervised learning.

Future Work

• Research Paper: Currently preparing a comprehensive research paper detailing the methodology, results, and potential future improvements.
• Improved Dataset: Plan to incorporate more diverse datasets for broader applicability.
• Real-Time Implementation: Extend the system to handle real-time video feeds for live surveillance applications.
• Enhanced Models: Explore advanced architectures such as Vision Transformers (ViT) or hybrid models for better performance.

Getting Started

Prerequisites

• Python
• TensorFlow
• OpenCV
• NumPy
• Pandas
• Matplotlib

Installation

2. Install dependencies:

   pip install -r requirements.txt

3. Download and preprocess the UCF Crime Dataset.