Edge-Optimized AI Surveillance Pipeline with Active Voice Alerting for High-Value Showrooms

This project underscores the immense practical advantages of combining artificial intelligence with edge computing for secure, low-latency surveillance. The integration of IMU data and RGB streams vastly outperforms traditional CCTV setups in anomaly detection, while processing data at the edge inherently solves the communication delays and privacy vulnerabilities associated with external cloud dependency. Furthermore, the addition of automated voice warnings elevates the system from a passive observer to an active security participant. Future extensions of this work could integrate adaptive learning systems, incorporate larger multimodal datasets, and utilize advanced transformer architectures to deploy this framework across wider, more complex commercial environments beyond retail showrooms.

1.         B. Solmaz et al., "Identifying behaviors in crowd scenes using stability analysis," IEEE Trans. Pattern Anal. Mach. Intell., 2012. 2.         E. Kim et al., "Ultra-low latency deterministic networking using SDN and NETCONF," IEEE Commun. Mag., 2018. 3.         X. Zhang et al., "Efficient surveillance video coding using background reconstruction," IEEE Trans. Circuits Syst. Video Technol., 2012. 4.         F. Esparza-Segundo et al., "ESP32-CAM based smart surveillance with Telegram integration," 2025. 5.         M. R. Palav et al., "AI-IoT framework for real-time customer behavior analysis," 2024. 6.         S. Kanagamalliga et al., "Real-time telephonic speech transcription and synthesis system," 2024. 7.         C. L. Fabian et al., "Machine learning approaches for multimodal Human Activity Recognition," 2024. 8.         G. A. Martinez-Mascorro et al., "Suspicious activity detection using 3D CNNs," 2020. 9.         L. Wang et al., "Multimodal few-shot learning for behavioral analysis," 2024. 10.     A. Ghadiya et al., "Cross-modal fusion for weakly supervised video anomaly detection," 2024.