Skip to content

yuvch98/FindingAnomalyInVideoAnalysis

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Video Anomaly Detection Analysis

This project is focused on developing a machine learning model for anomaly detection in video data. The main objective is to classify video events as either normal or anomalous, which is crucial in various applications like security surveillance, safety monitoring, and event detection systems.

Overview

The code provides a comprehensive workflow for video anomaly detection, starting from data preparation to model training, validation, and testing. It includes the following key components:

Data Preparation

The dataset consists of video files categorized into normal and anomalous events, sourced from Google Drive. Anomalous events are further divided into specific types such as Abuse, Arson, Assault, etc., while normal events are grouped under Normal categories. Data is split into training and testing sets, ensuring a balanced representation of all event types.

Preprocessing

Video frames are converted to grayscale and resized to a standard dimension for uniformity. A Structural Similarity Index Measure (SSIM) is used to identify frames with significant changes, termed as "interest frames", which are likely to indicate an anomaly. Preprocessed data is then prepared for model input, including the application of data augmentation techniques for training robustness.

Model Architecture

A 3D Convolutional Neural Network (CNN) model is employed, enhanced with spatial attention mechanisms to focus on relevant spatial features within video frames. The model architecture includes convolutional layers, pooling layers, batch normalization, and fully connected layers, concluding with a sigmoid activation function for binary classification.

Training and Evaluation

The model is trained using the preprocessed training dataset, with validation performed on a separate validation set to monitor overfitting. Performance is evaluated on the test dataset, with metrics such as accuracy, loss, and a confusion matrix providing insight into model effectiveness. Various callbacks like ModelCheckpoint and EarlyStopping are implemented to enhance training by saving the best model and preventing overfitting, respectively.

Purpose

The project aims to automate the detection of anomalous events in video streams, which can significantly enhance the efficiency and effectiveness of surveillance and monitoring systems. By leveraging deep learning and attention mechanisms, the model can learn to identify intricate patterns and anomalies in diverse video contexts.

Usage

To utilize this model for anomaly detection:

Prepare your video dataset, categorizing videos into normal and anomalous events. Ensure the dataset follows a similar structure as described in the Data Preparation section. Train the model using your dataset, adjusting hyperparameters and model architecture as necessary for optimal performance. Deploy the trained model to analyze new video data, classifying events based on the learned patterns.

Conclusion

This Video Anomaly Detection project demonstrates the power of deep learning in analyzing and understanding complex video data. The developed model can serve as a foundation for various applications requiring automated event detection and anomaly identification in video streams.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages