Behavior and Zone Analytics System : Entity Action Monitoring

Sim, Hor Kang (2026) Behavior and Zone Analytics System : Entity Action Monitoring. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text RSW_Sim Hor Kang_Fulltext.pdf Restricted to Registered users only Download (4MB)

Abstract

This project explores the development of an AI-powered surveillance system that integrates object detection, tracking, and behavioral analysis using deep learning techniques. The primary goal is to offer an open, scalable alternative to proprietary surveillance software dominating the market, which often restricts adaptability and customization. By leveraging real-time analytics and reducing the cognitive workload on human operators, the system transforms raw video feeds into meaningful insights to enhance situational awareness and security responses. The system uses YOLO for real-time human detection and StrongSORT for multi-object tracking enabling the identification of anomalous or predefined actions within monitored zones. The project adopts an Iterative Development Model, allowing continuous refinement of system requirements and performance. Core modules include the user interface, image transformation, human detection, tracking, and behavior analytics. Functional and non-functional requirements were defined early to ensure the system meets usability, performance, security, and reliability standards. Various tools and frameworks, including OpenCV and PyTorch, were employed to construct and test the AI modules while Node.js and embedded JavaScript were used to create the web interface. A total of 15 test cases were executed to validate the accuracy of the detections In summary, this project demonstrates the feasibility of using AI to automate video surveillance tasks while providing a cost-effective and adaptable alternative to existing solutions. While certain limitations were identified such as fixed output resolution and overuse of system resources, the overall system proved functional and impactful. Future improvements will focus on expanding behavior datasets, enhancing recognition accuracy in low-light or occluded environments, and supporting deployment across edge devices for scalable adoption.