Tunnel Eye : AI Based Visual Inspection For Refinery Pipelines

This paper presented the design and implementation of an IoT-based facial authentication system for secure and intelligent door access control. The proposed system integrates facial recognition, real-time monitoring, wireless communication, and automated door control into a single embedded platform. The experimental results demonstrate that the system is capable of reliably detecting the presence of a person, capturing images, and performing facial recognition for identity verification. The integration of the ESP32-CAM module enables efficient image processing and communication, while the use of Wi-Fi connectivity allows real-time notification and remote access control. The system successfully allows the authorized user to verify visitors and make decisions to grant or deny access, ensuring improved security and convenience. The proposed solution offers a cost-effective, scalable, and efficient approach to modern security systems, reducing dependency on traditional access methods such as keys and passwords. The combination of sensing, processing, and communication enhances system performance and user experience. However, certain limitations such as dependency on network connectivity, lighting conditions, and recognition accuracy may affect system performance. Future improvements can focus on enhancing facial recognition accuracy, improving system security, and integrating additional features such as cloud storage. In conclusion, the developed system demonstrates the practical implementation of IoT and facial recognition technology for smart security applications and provides a strong foundation for further research in intelligent access control systems

[1]         Lin, J., Zhu, L., Chen, W.-M., Wang, W.-C., & Han, S. (2024). Tiny    machine                    learning:        Progress        and       futures. https://arxiv.org/abs/2403.19076. Use for: TinyML overview and why on-device models matter. [2]         George, A., Ecabert, C., Otroshi Shahreza, H., Kotwal, K., & Marcel, S. (2023). EdgeFace: Efficient face recognition model for edge devices. https://arxiv.org/abs/2307.01838. Use for: Lightweight face-recognition model design for constrained hardware. facerecognition using MobileNetV2 and FaceNet deep-learning implementations Use for: Practical IoT implementation deployment. [3]         Elhanashi, A. (2024). Advancements in TinyML: Applications, limitations, and impact on IoT devices. Electronics (MDPI), 13(17), Article 3562. https://doi.org/10.3390/electronics13173562 Use for: TinyML tools, deployment challenges and device constraints.   [4]         Heydari, S., et al. (2024). Tiny machine learning and on-device inference: A survey. PMC  preprint. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12115890   [5]         Xie, Y., Li, J., & Chen, X. (2022).Privacy protection framework for face recognition in edge-based Internet of Things. Future Generation Computer Systems (conference/journal entry [6]         Al-Nabulsi, J., & Others. (2023).IoT solutions and AI-based frameworks for masked-face recognition. Sensors (MDPI), 23(16), 7193. https://doi.org/10.3390/s23167193Use for: Masked-face recognition methods and system design.   [7]         Eman, M., et al. (2023).Innovative hybrid approach for masked face recognition using SSD-MobileNetV2 and RPCA. Sensors (MDPI), 23(15), 6727.https://doi.org/10.3390/s23156727Use for: Practical model strategies for occluded/masked faces . [8]         Wang, Y., et al. (2023). Masked face recognition system based on attention mechanism. Information (MDPI),14(2), Article 87.https://doi.org/10.3390/info14020087 Use for: Attention-based architectures to improve masked F R [9]         Kumar, B. A., et al. (2023).Face mask detection on photo and real-time video with MobileNetV2 variants. Applied Sciences / Computers. https://www.mdpi.com/2076-3417/13/2/935Use for: MobileNetV2 adaptations for real-time mask detection on embedded platforms. [10]      Mao, Y., Yu, X., Huang, K., Zhang, Y.-J. A., & Zhang, J. (2023). Green edge AI: A contemporary survey. https://arxiv.org/abs/2312.00333 Use for: Energy considerations and efficient edge AI design.