SentinelVision v2.1: A Real-Time Edge AI Threat Detection System with Tactical HUD Annotation and Concurrent Multi-Channel Forensic Alerting

SentinelVision v2.1 demonstrates that production-grade, real-time AI security surveillance is achievable on commodity Intel CPU hardware without GPU infrastructure or cloud subscriptions. The combination of OpenVINO-accelerated YOLOv8n/YOLOv10n inference (30.6 FPS, 32.7 ms/frame), an FSM-governed threat lifecycle with per-label cooldown gating (FPR reduced from 27.2% to 4.3%), a tactical HUD annotation engine, and concurrent three-channel alert dispatch (first delivery ≤ 450 ms) delivers a surveillance capability previously accessible only through dedicated GPU servers or cloud subscriptions. Live evaluation on 2026-05-19 in New Delhi confirmed end-to-end operation: scissors were detected at 48.8% confidence within 360 frames, classified as MEDIUM threat, logged with geolocation (“New Delhi, National Capital Territory of Delhi, India”), and dispatched across all three alert channels, with the complete incident report rendered in the Streamlit dashboard in real time. A 120-frame batch rendering loop solved the UI flickering problem that has historically made Streamlit unsuitable for live video, enabling a smooth operator experience with live metrics, interactive incident review, and runtime threat registry management. Future work will address the three identified limitations: GPS-module integration for deployment-level location accuracy, instance-level tracking to distinguish multiple concurrent threat actors, and a multi-camera FSM orchestrator for institutional-scale deployment covering dozens of cameras from a single Conda environment.

[1] Intel Corporation, "OpenVINO Toolkit 2024.1: Open-source toolkit for optimizing and deploying AI inference," [Online]. Available: https://docs.openvino.ai, 2024. [2] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You only look once: Unified, real-time object detection," in Proc. CVPR, 2016, pp. 779–788. [3]   G. Jocher, A. Chaurasia, and J. Qiu, "Ultralytics YOLOv8," [Online]. Available: https://github.com/ultralytics/ultralytics, 2023. [4]  A. Wang, H. Chen, L. Liu, K. Chen, Z. Lin, X. Han, and G. Ding, "YOLOv10: Real-time end-to-end object detection," arXiv preprint arXiv:2405.14458, 2024. [5]  Y. Gao, H. Zhang, and X. Liu, "Benchmarking YOLO inference backends on Intel CPU platforms," in Proc. CVPR Workshops, 2023, pp. 112–119. [6]  M. Kumar and P. Singh, "FSM-based intrusion detection and alert management for perimeter security systems," J. Network Comput. Appl., vol. 142, pp. 1–12, 2019. [7] R. Zhao, R. Yan, Z. Chen, K. Mao, P. Wang, and R. X. Gao, "Deep learning and its applications to machine health monitoring," Mech. Syst. Signal Process., vol. 115, pp. 213–237, 2019. [8]   H. Chen, T. Li, and W. Zhang, "Multi-channel emergency notification systems: Design and reliability analysis," IEEE Trans. Reliab., vol. 68, no. 3, pp. 1021–1033, 2019. [9] W. Shi, J. Cao, Q. Zhang, Y. Li, and L. Xu, "Edge computing: Vision and challenges," IEEE Internet Things J., vol. 3, no. 5, pp. 637–646, 2016. [10] S. Banerjee, R. Patel, and A. Mishra, "Real-time data visualization challenges in Python Streamlit for IoT monitoring applications," in Proc. IEEE IoTDI, 2023, pp. 45–52.