Multi streams face recognition system for student tracking

Chow, Mun Kent (2025) Multi streams face recognition system for student tracking. Final Year Project, UTAR.

PDF Download (5Mb)

Abstract

With the increasing demand for efficient space utilization and data-driven management in academic libraries, accurate and real-time occupancy monitoring has become an essential requirement. Traditional manual people counting methods are labour-intensive, error-prone, and incapable of providing continuous or historical insights for operational planning. To address these limitations, this project proposes an intelligent computer vision–based people counting and occupancy analytics system specifically designed for library environments. The proposed system utilizes a real-time video processing pipeline that integrates YOLOv8 for human detection with ByteTrack for multi-object tracking. A strict zone-transition–based counting mechanism is implemented to accurately detect entry and exit events while minimizing double counting and false detections. Multiple anti-duplicate strategies, including spatial suppression, temporal cooldown, and minimum track age validation, are applied to improve robustness under real-world conditions such as occlusion, lighting variation, and dense crowd movement. Experimental results demonstrate that the system achieves approximately 95% counting accuracy in practical deployment scenarios. Beyond real-time monitoring, the system incorporates a structured database architecture to support comprehensive historical analysis and predictive analytics. Occupancy data is aggregated at hourly, daily, weekly, and monthly levels, enabling peak-hour identification, trend analysis, and performance evaluation. Predictive analytics based on historical patterns and day-of-week analysis are implemented to forecast future occupancy levels and generate capacity alerts, transforming the system into an effective decision-support tool for library management. The system is designed with a modular and scalable multi-camera architecture, allowing independent processing of multiple locations while providing a unified web-based dashboard for visualization. The dashboard presents live occupancy, historical trends, statistical summaries, and predictive insights through an intuitive interface accessible to non-technical users. By combining modern computer vision techniques, robust system architecture, and data analytics, this project demonstrates a practical and privacy-conscious solution for intelligent library occupancy monitoring and management.

Actions (login required)