Dress Code Monitoring Method in Industrial Scene Based on Improved YOLOv8n and DeepSORT.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Deep learning object detection is vital for dress code monitoring but struggles with small targets, leading to false alarms and missed detections.
• Existing state-of-the-art models require improvement to handle the nuances of real-world dress code enforcement, especially concerning accessories.

Purpose of the Study:

• To propose a novel and improved deep learning method for accurate dress code monitoring.
• To enhance object detection capabilities for small targets like hats and masks.
• To reduce false alarms and missed detections in real-world dress code enforcement scenarios.

Main Methods:

• An improved YOLOv8n model incorporating a novel FPN-PAN-FPN (FPF) neck structure for feature fusion.
• Integration of Receptive-Field Attention convolutional operation (RFAConv) and Focused Linear Attention (FLatten) mechanism for enhanced feature extraction and receptive field expansion.
• Implementation of DeepSORT tracking for multi-frame instance information and a new dress code judgment criterion for real-scene monitoring.

Main Results:

• The improved YOLOv8n model demonstrated increased mean Average Precision (mAP) while reducing model size.
• The integrated system effectively identified dress code violations in real-world scenarios.
• The proposed method significantly reduced false alarms and missed detections compared to baseline models.

Conclusions:

• The novel method offers a robust solution for automated dress code monitoring, improving accuracy and reliability.
• The enhancements to the YOLOv8n model and the new judgment criterion address key limitations in detecting small objects.
• This approach provides a practical and effective tool for real-world applications requiring precise object detection and classification.