AI-Based Real-Time Person Counting and Security Monitoring System Using YOLO Object Detection and Multi-Object Tracking

AbstractManual counting of people at entrances, exhibitions and access-controlled premises is slow, labour-intensive anderror-prone, motivating automated, vision-based alternatives. This paper presents the design, implementation and fieldtesting of an AI-based counter and security monitoring system that couples a deep-learning object detector (YOLO)with a multi-object tracker to count and monitor people crossing a virtual reference line in live video. Frames capturedfrom a CCTV/webcam feed are processed with OpenCV; a pretrained YOLO model localises every person in theframe, and a tracking stage assigns each detection a persistent identity so that a directional line-crossing eventincrements or decrements a running occupancy count exactly once per individual. The system was built around aGPU-accelerated workstation and a high-frame-rate camera, and was validated through indoor laboratory trials,deployment at the department's administration block, and a public demonstration at the “Vignana Mela 4.0” exhibition(17–19 February 2025). During these trials the system tracked multiple individuals simultaneously with detectionconfidences above 0.85, distinguished moving people from static background objects such as furniture withoutregistering false counts, and sustained continuous operation while accumulating counts beyond ten thousand crossingswithout loss of tracking continuity. The results indicate that the proposed pipeline is a practical, low-cost and scalablealternative to manual headcounting for occupancy management, crowd-density estimation and security monitoring ininstitutional, commercial and public spaces. Directions for future work, including multi-camera fusion, edgedeployment and cloud-based analytics, are also discussed.