Intelligent CCTV Monitoring on Campus Using YOLOv8 and Deep SORT: Design and Prototype

Abstract

This paper presents the design and prototype of a smart campus surveillance system that performs real-time person detection and multi-object tracking on CCTV feeds by pairing YOLOv8 (detector) with Deep SORT (tracker). The pipeline assigns stable IDs across frames, logs movement to a lightweight CSV store, and aggregates trajectories into movement heatmaps to reveal spatial utilization. A Streamlit dashboard integrates the live annotated feed with operator-oriented analytics (people counts, recent logs, and heatmap views), enabling situational awareness without altering existing CCTV infrastructure. Architectural choices emphasize on-premises processing (privacy, low latency), modularity, and operator usability. The prototype’s software design, deployment considerations, and a planned evaluation protocol (throughput and qualitative tracking stability) are detailed, while exhaustive quantitative metrics are deferred to a subsequent journal article. The system targets single-camera operation per pipeline with per-camera analytics; cross-camera re-identification and event detection are identified as future work. Overall, the contribution provides a reproducible blueprint for campus environments with constrained resources and strict privacy requirements.