Date of Award

12-2024

Document Type

Project

Degree Name

Master of Science in Information Systems and Technology

Department

Information and Decision Sciences

First Reader/Committee Chair

Conrad, Shayo

Abstract

This culminating experience project explored innovative methods for enhancing anomaly detection in video surveillance systems, a vital concern for public safety and security management. The research questions addressed were: Q1) What emergent approaches in Activity-based Human Action Recognition (AbHAR) lead to significant advancements in surveillance technology driven by human cognition? Q2) How can the processing and detection phases of video monitoring systems be optimized for improved efficiency? Q3) What are the advantages of ensemble approaches over individual algorithms in enhancing the robustness and accuracy of anomaly detection systems in video surveillance?

The findings were: (Q1), That deep learning methodologies, particularly Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks, significantly elevate the ability to recognize and interpret human actions, thereby improving anomaly detection systems. Compared to traditional machine learning methods, CNNs demonstrate improved accuracy in spatial feature extraction, by 50% and LSTM networks perform better in detecting sequential patterns with 51% accuracy, underscoring the advantages of deep learning in enhancing AbHAR-based systems (Q2) Integrating edge computing with advanced algorithms significantly enhances the efficiency of video processing and anomaly detection, reducing latency by 30% and increasing real-time processing capability by 20%, which is crucial for effective surveillance operations. This combination allows the system to process data closer to the source, thus enabling faster decision-making and reducing bandwidth usage. Additionally, by optimizing processing at the edge, the system can better handle high volumes of video data with improved accuracy and speed, essential for timely threat detection and response. (Q3) The study underscored the benefits of ensemble methods, which combine multiple algorithms to improve detection accuracy by 10% and reduce false positives, thereby optimizing the response to security threats. By leveraging diverse algorithmic strengths, ensemble methods provide a more robust system that adapts to varied surveillance environments and evolving anomaly patterns. This approach not only enhances overall reliability but also supports proactive threat management, ensuring more accurate and timely security interventions.

The conclusions were: (Q1) deep learning models significantly enhance action recognition capabilities, (Q2) edge computing optimizes processing for timely anomaly detection and (Q3) ensemble techniques improve system robustness and accuracy. Areas for further study include: (1). investigating multi-modal data integration to enrich contextual analysis, (2). leveraging innovative learning paradigms to reduce dependency on labeled datasets, and (3). refining algorithms for effective real-time applications across diverse surveillance settings.

Download

  Contact Author

Share

COinS