Related Experiment Video
Updated: Aug 24, 2025
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
9.1K
Efficient anomaly recognition using surveillance videos
Gulshan Saleem1, Usama Ijaz Bajwa1, Rana Hammad Raza2
1Department of Computer Science, COMSATS University Islamabad, Lahore Campus, Lahore, Pakistan.
Peerj. Computer Science
|October 20, 2022
Summary
This study introduces an efficient and cost-effective smart surveillance system for anomaly recognition. The Temporal based Anomaly Recognizer (TAR) framework improves accuracy and reduces computational costs in real-time video analysis.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Smart surveillance is crucial for human safety but faces challenges in cost, real-time processing, and accurate anomaly detection.
- Existing systems, especially those using 3D convolutional neural networks, are computationally expensive and impractical for widespread deployment.
- Automated surveillance struggles with defining anomalies and processing large volumes of data like 24/7 CCTV footage.
Purpose of the Study:
- To develop a resource-efficient and cost-effective framework for anomaly recognition in smart surveillance.
- To address the computational overhead bottleneck in real-time automated surveillance systems.
- To achieve high accuracy in anomaly recognition while minimizing computational resource requirements.
Main Methods:
- Proposed the Temporal based Anomaly Recognizer (TAR) framework, combining a partial shift strategy with a 2D convolutional architecture (MobileNetV2).
- Evaluated the framework on the UCF Crime dataset for anomaly recognition.
- Tested the model's performance on spatially augmented surveillance videos for both two-class and multi-class anomaly recognition.
Main Results:
- Achieved 88% accuracy on the UCF Crime dataset, outperforming state-of-the-art by 2.47%.
- Obtained 52.7% accuracy for multi-class anomaly recognition on the UCF Crime2Local dataset.
- Demonstrated real-time capability by handling six simultaneous camera streams without additional resources.
Conclusions:
- The TAR framework offers an efficient and accurate solution for anomaly recognition in smart surveillance.
- The proposed system is cost-effective and suitable for real-time applications, overcoming previous computational limitations.
- This research advances the practical implementation of smart surveillance for enhanced public and private safety.
