Mask Detect and Counting

https://www.youtube.com/watch?v=dz6mcj71wLc

A real-time mask detection and counting system based on YOLOv8s that identifies mask-wearing status in live CCTV footage and counts the total number of people and mask-wearing individuals through object tracking.

project structure

Mask-detect-and-counting/
├── mask_checking/         # Training results and model files
├── data.yaml              # Data path and class name configuration
├── hyp_custom.yaml        # YOLOv8s training configuration
├── main.ipynb             # YOLOv8s fine-tuning Jupyter notebook
├── mask_counting.py       # Main detection & counting
├── yolov8s.pt             # pretrained model
└── README.md

Main Functions

• YOLOv8s training in Jupyter notebook
• Detecting masks, assigning identity IDs, and tracking
• Counting masked and non-masked people

Training Result

Metric	Value	Description
Precision	88.89%	Accuracy of positive predictions
Recall	87.21%	Ability to find all positive instances
mAP50	91.73%	Mean Average Precision at IoU 0.5
mAP50-95	59.71%	Mean Average Precision at IoU 0.5-0.95

Results are from the final epoch (epoch 100) with best model weights.

Model infomation

• Model: YOLOv8s (Ultralytics)
• Classes:
  • 0: Face with mask
  • 1: Face without mask
• Training Data: OpenCV mask image dataset
• Framework: Ultralytics YOLO

Limitations

• Object detection fails when object size is too small, resulting in inaccurate counting
• Individual tracking is based on bounding box positions, making it unable to distinguish between people within the same bounding box

---

📎 Reference Libraries

• Ultralytics YOLO
• abewley/sort