Robustness and Optimization of Mobile Facial Recognition Models On Edge Devices Under Real-World Conditions

• Author(s): Agbo Nnamdi Vitus ; Ochi Victor Chukwudi
• Paper ID: 1710706
• Page: 893-897
• Published Date: 18-09-2025
• Published In: Iconic Research And Engineering Journals
• Publisher: IRE Journals
• e-ISSN: 2456-8880
• Volume/Issue: Volume 9 Issue 3 September-2025

Abstract

This paper evaluates the robustness and optimization of real-time facial recognition models deployed on mobile edge devices under varying environmental conditions. With the growing reliance on mobile platforms for authentication, surveillance, and security applications, ensuring reliable performance across diverse scenarios is critical. We benchmark lightweight deep learning models including MobileNetV2, FaceNet, and EfficientNet on Android devices, analyzing trade-offs between accuracy, latency, and energy consumption. Experiments were conducted under controlled variations in illumination, background complexity, and face orientation to assess model robustness. The results demonstrate that while MobileNetV2 offers superior efficiency with reduced computational overhead, FaceNet provides higher recognition accuracy at the expense of increased latency and battery usage. Our findings emphasize the importance of balancing performance and efficiency when deploying face recognition systems on edge devices. This study contributes to the ongoing optimization of mobile AI for real-world use cases such as mobile security, examination authentication, and smart surveillance.

Keywords

Mobile AI, Edge Computing, Facial Recognition, Model Optimization, Real-Time Systems, Environmental Conditions, Android Devices

Citations

IRE Journals:
Agbo Nnamdi Vitus , Ochi Victor Chukwudi "Robustness and Optimization of Mobile Facial Recognition Models On Edge Devices Under Real-World Conditions" Iconic Research And Engineering Journals Volume 9 Issue 3 2025 Page 893-897

IEEE:
Agbo Nnamdi Vitus , Ochi Victor Chukwudi "Robustness and Optimization of Mobile Facial Recognition Models On Edge Devices Under Real-World Conditions" Iconic Research And Engineering Journals, 9(3)