This paper presents a real-time security guard uniform recognition system designed for surveillance environments using edge computing. The proposed approach combines U-Net for precise cloth segmentation and MobileNetV2 for lightweight uniform classification, optimized to run efficiently on the NVIDIA Jetson Nano. The system segments key uniform components such as shirts, pants, belts, caps, shoes, and ties, then classifies them to distinguish security uniforms from regular clothing. Trained on the DeepFashion2 dataset and a custom uniform dataset, the model achieves high accuracy, precision, recall, and F1-score while maintaining real-time performance (10–15 FPS) on resource-constrained hardware. The results demonstrate that integrating segmentation with classification significantly improves detection reliability, making the system suitable for scalable, privacy-preserving, and automated security monitoring applications.
Goyal et al. (Thu,) studied this question.