Smart Vision Systems with Edge AI: Semiconductor-Based Architectures for Real-Time Facial Recognition and Object Detection
DOI:
https://doi.org/10.63282/3050-922X.IJERET-V7I1P134Keywords:
Edge AI, Smart Vision Systems, Semiconductor Accelerators, Real-Time Computer Vision, Facial Recognition, Object Detection, Embedded AI Systems, Edge Computing, Deep Learning InferenceAbstract
Smart vision systems have become central to modern digital infrastructure, supporting applications such as intelligent surveillance, autonomous mobility, industrial inspection, and smart retail analytics. Many existing computer vision systems rely on cloud-based processing, where image data is transmitted to remote servers for model inference. While this approach enables large-scale computation, it introduces latency, bandwidth dependency, and concerns related to privacy and data governance. Recent developments in Edge AI provide an alternative paradigm in which inference is performed directly on local devices using optimized neural networks and specialized hardware accelerators.
This study examines the role of semiconductor-based architectures in enabling real-time facial recognition and object detection within edge-deployed vision systems. The paper presents a conceptual framework that integrates image sensors, embedded processors, and dedicated AI accelerators designed to support efficient deep learning inference at the device level. The proposed architecture emphasizes parallel processing, optimized memory pipelines, and lightweight model deployment to support low-latency visual analytics. Performance considerations such as inference latency, throughput, energy consumption, and detection accuracy are examined through a comparative evaluation of different deployment platforms, including cloud GPU systems, edge CPUs, and semiconductor-assisted edge accelerators.
The analysis indicates that semiconductor-assisted edge architectures substantially reduce inference latency while improving computational efficiency compared with conventional cloud-based pipelines. In addition to performance gains, localized inference reduces the need to transmit sensitive visual data to external infrastructure, which strengthens privacy protection and improves system reliability in environments with limited connectivity. The findings highlight the growing importance of semiconductor-enabled Edge AI as a foundation for scalable and responsive smart vision systems across multiple sectors
References
[1] Alam, S., Chowdhury, M., & Hasan, M. (2024). Survey of deep learning accelerators for edge and embedded computing. Electronics, 13(15), 2988.
[2] Mittal, P., Singh, A., & Kaur, R. (2024). A comprehensive survey of deep learning-based lightweight object detection models for edge devices. Artificial Intelligence Review, 57, 10877.
[3] Wang, X., Liu, Y., & Chen, L. (2025). A comprehensive survey on on-device AI models: Design, optimization, and deployment. ACM Computing Surveys.
[4] Gao, P., Chen, H., & Zhang, W. (2025). Emerging electronic technologies enabling next-generation AI systems. Intelligent and Converged Networks.
[5] Amirgaliyev, B., Ziyatdinov, A., & Omarov, B. (2025). A review of machine learning and deep learning methods for face recognition and person detection. IEEE Access.
[6] Surantha, N., Rahman, F., & Nugroho, H. (2025). Key considerations for real-time object recognition on edge computing platforms. Applied Sciences, 15(13), 7533.
[7] Yuan, Q., Zhao, Y., & Ahmad, J. (2026). Dual-engine embedded face detection and recognition framework using YOLO-based architecture. Informatica.
[8] Vasile, C. E., Dumitrescu, A., & Popescu, M. (2024). Image processing hardware acceleration: A comprehensive review of architectures and platforms. Sensors.
[9] Yang, Y., Kneip, A., & Frenkel, C. (2024). EvGNN: An event-driven graph neural network accelerator for edge vision. IEEE Transactions on Circuits and Systems for Video Technology.
[10] Tain, B., Millet, R., Lemaire, R., et al. (2025). J3DAI: A tiny deep neural network-based accelerator for 3D stacked CMOS image sensors. IEEE Journal of Solid-State Circuits.
[11] Nag, S., Datta, G., Kundu, S., Chandrachoodan, N., & Beerel, P. A. (2023). ViTA: A vision transformer inference accelerator for edge applications. IEEE Transactions on Very Large Scale Integration Systems.
[12] Montgomerie-Corcoran, A., Toupas, P., Yu, Z., & Bouganis, C. (2023). SATAY: A streaming architecture toolflow for accelerating YOLO models on FPGA devices. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems.
[13] Zhou, K., Zhang, Y., Ren, S., & Sun, J. (2021). Deep learning for object detection: A survey. Computer Vision and Image Understanding, 203, 103107.
[14] Bochkovskiy, A., Wang, C. Y., & Liao, H. Y. M. (2020). YOLOv4: Optimal speed and accuracy of object detection. arXiv preprint arXiv:2004.10934.
[15] Redmon, J., & Farhadi, A. (2020). YOLOv3: An incremental improvement. arXiv preprint arXiv:1804.02767.
[16] Tan, M., Pang, R., & Le, Q. V. (2020). EfficientDet: Scalable and efficient object detection. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.
[17] Howard, A., Sandler, M., Chu, G., et al. (2020). Searching for MobileNetV3. Proceedings of the IEEE/CVF International Conference on Computer Vision.
[18] Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., & Chen, L. C. (2020). MobileNetV2: Inverted residuals and linear bottlenecks. IEEE Conference on Computer Vision and Pattern Recognition.
[19] Deng, J., Guo, J., Niannan, X., & Zafeiriou, S. (2020). ArcFace: Additive angular margin loss for deep face recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence.
[20] Schroff, F., Kalenichenko, D., & Philbin, J. (2020). FaceNet: A unified embedding for face recognition and clustering. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.
[21] Chen, T., Moreau, T., Jiang, Z., et al. (2021). TVM: An automated end-to-end optimizing compiler for deep learning. USENIX Symposium on Operating Systems Design and Implementation.
[22] Alampally, J. (2022). Designing High-Performance OLAP Cubes for Advanced Analytical Decision-Making. Frontiers in Computer Science and Artificial Intelligence, 1(1), 31-36.
[23] ALAMPALLY, J. (2022). Prescriptive analytics on anonymized patient data using regression and distributed computing. Journal of Computer Science and Technology Studies, 4(1), 107-111.
[24] Jagadeeswar, A. Optimizing Enterprise BI Platforms for High-Volume Healthcare Data Warehouses. J Artif Intell Mach Learn & Data Sci 2021, 4(2), 3270-3273.
[25] Satyanarayanan, M. (2020). The emergence of edge computing. Computer, 50(1), 30–39.
[26] Shi, W., Cao, J., Zhang, Q., Li, Y., & Xu, L. (2021). Edge computing: Vision and challenges. IEEE Internet of Things Journal, 3(5), 637–646.
[27] Li, Y., Ota, K., & Dong, M. (2021). Deep learning for smart industry: Efficient vision systems on edge devices. IEEE Communications Surveys & Tutorials.
[28] Zhou, Z., Chen, X., Li, E., Zeng, L., Luo, K., & Zhang, J. (2020). Edge intelligence: Paving the last mile of artificial intelligence with edge computing. Proceedings of the IEEE, 107(8), 1738–1762.
[29] Zhang, C., Li, P., Sun, G., Guan, Y., Xiao, B., & Cong, J. (2021). Optimizing FPGA-based accelerator design for deep convolutional neural networks. Proceedings of the ACM/SIGDA FPGA Conference.
[30] Sze, V., Chen, Y. H., Yang, T. J., & Emer, J. (2020). Efficient processing of deep neural networks: A tutorial and survey. Proceedings of the IEEE, 105(12), 2295–2329.
[31] Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C. Y., & Berg, A. C. (2020). SSD: Single shot multibox detector. European Conference on Computer Vision.
