Optimized Culprit Identification Using Mobilenet and Attention Mechanisms

Researchers propose an optimized deep learning model combining MobileNet with attention mechanisms for automated facial identification in surveillance systems, achieving 97.8% accuracy while maintaining computational efficiency for real-time deployment.

This research addresses a fundamental challenge in computer vision: deploying high-accuracy facial recognition systems on resource-constrained hardware without sacrificing real-world performance. The integration of attention mechanisms with MobileNet represents a practical advancement in edge AI deployment, where balancing accuracy and computational burden remains critical.

The broader context reflects growing demand for efficient AI inference across surveillance infrastructure globally. As security systems expand, organizations increasingly require solutions that run locally on edge devices rather than relying on cloud processing, which introduces latency and privacy concerns. MobileNet architectures have emerged as industry standard for such applications, and this work's addition of selective attention mechanisms addresses a known limitation: standard lightweight models often sacrifice feature discrimination to achieve speed.

The technical improvements—97.8% accuracy outperforming baseline CNN, ResNet, and standard MobileNet—suggest meaningful practical gains for deployment scenarios. Testing across multiple datasets (LFW, CASIA-WebFace, VGGFace2) under realistic conditions (illumination variation, pose changes, occlusion) indicates robustness beyond laboratory conditions. The maintained low computational complexity positions this approach for edge deployment, where inference speed directly impacts operational costs and response times.

For the AI and surveillance technology sectors, this work validates that selective attention mechanisms can cost-effectively improve lightweight model performance. Organizations implementing edge-based security systems may find such optimized architectures reduce infrastructure requirements while improving accuracy. The research suggests continued evolution toward efficient models rather than reliance on increasingly larger models, influencing development priorities across computer vision applications.

• →MobileNet with attention mechanisms achieves 97.8% facial recognition accuracy, outperforming larger baseline models
• →The framework maintains computational efficiency suitable for real-time edge deployment in surveillance systems
• →Channel and spatial attention mechanisms selectively focus on discriminative features while suppressing background noise
• →Testing on multiple benchmark datasets under realistic conditions demonstrates robustness to illumination, pose, and occlusion variations
• →Reduced inference time and computational complexity make the approach practical for resource-constrained surveillance infrastructure