harleensachdev/yolo-lanyard-object-detector

YOLOv8 Lanyard Detector

A YOLOv8 nano model fine-tuned to detect ID cards/lanyards worn by people. This is Stage 2 of a two-stage detection pipeline for access control and security monitoring.

Model Description

• Model: YOLOv8n (nano) - optimized for speed and efficiency
• Task: Object Detection (Single Class)
• Class: id_tag (lanyard/ID card detection)
• Use Case: Security access control, attendance monitoring, event management

Two-Stage Detection Pipeline

This model is designed to work in a two-stage pipeline:

Stage 1: Person Detection

• Use YOLOv8 COCO pretrained model
• Detects people in the scene

Stage 2: Lanyard Detection (This Model)

• Runs on the full image
• Detects lanyards/ID cards
• Output: Person wearing lanyard ✓ or No lanyard ✗

Decision Logic

• PASS: Both person AND lanyard detected → Person has valid ID
• FAIL: Person detected but NO lanyard → Access denied

Usage

Install Dependencies

pip install ultralytics opencv-python huggingface_hub

Quick Start

from ultralytics import YOLO
import cv2

# Load model
model = YOLO("harleensachdev/yolo-lanyard-detector")

# Run detection
image = cv2.imread("your_image.jpg")
results = model(image)

# Check if lanyard detected
has_lanyard = len(results[0].boxes) > 0

if has_lanyard:
    print("✓ Lanyard detected - Access granted")
else:
    print("✗ No lanyard - Access denied")

Two-Stage Detector Script

from ultralytics import YOLO
import cv2

# Stage 1: Detect people
person_detector = YOLO("yolov8n.pt")  # COCO pretrained
person_results = person_detector(image)

# Check if person detected (class 0 = person in COCO)
has_person = any(int(box.cls[0]) == 0 for box in person_results[0].boxes)

# Stage 2: Detect lanyard
lanyard_detector = YOLO("harleensachdev/yolo-lanyard-detector")
lanyard_results = lanyard_detector(image)
has_lanyard = len(lanyard_results[0].boxes) > 0

# Final decision
if has_person and has_lanyard:
    print("✓ PASS: Person with valid ID")
elif has_person and not has_lanyard:
    print("✗ FAIL: Person without ID")
else:
    print("⚠️  No person detected")

Model Performance

• Architecture: YOLOv8n
• Input Size: 640x640
• Parameters: ~3M (lightweight, real-time capable)
• Training Dataset: Custom lanyard detection dataset (908 images)
  • Positive examples: 837 images with lanyard annotations
  • Negative examples: 71 images without lanyards
  • Total annotations: 1,360 bounding boxes

Training Details

• Framework: Ultralytics YOLOv8
• Epochs: 100 (with early stopping)
• Batch Size: 16
• Image Size: 640x640
• Optimizer: Adam
• Data Augmentation:
  • Horizontal flip (50%)
  • Rotation (±10°)
  • Translation (±10%)
  • HSV augmentation
  • Mosaic augmentation

Intended Use

Primary Applications

• 🏢 Corporate Access Control: Monitor employee ID badge compliance
• 🎓 Educational Institutions: Verify student ID cards
• 🏥 Healthcare Facilities: Ensure staff identification
• 🎪 Events & Conferences: Attendee credential verification
• 🏭 Industrial Sites: Safety and security compliance

Use Cases

• Real-time video surveillance for access control
• Automated attendance systems
• Visitor management
• Security checkpoint automation
• Compliance monitoring

Limitations

• Optimized for frontal/side views of people
• Performance may degrade with:
  • Extreme lighting conditions
  • Heavy occlusion of lanyard
  • Very small or distant subjects
  • Non-standard ID badge formats

Ethical Considerations

This model is designed for security and access control purposes within the framework of SDG 16 (Peace, Justice and Strong Institutions). Users should:

• ✓ Ensure compliance with local privacy laws
• ✓ Obtain proper consent when deploying in surveillance systems
• ✓ Use only in authorized security contexts
• ✗ Avoid discriminatory applications
• ✗ Do not use for unauthorized surveillance

Citation

@software{yolov8_lanyard_detector,
  author = {Harleen Sachdev},
  title = {YOLOv8 Lanyard Detector},
  year = {2026},
  publisher = {HuggingFace},
  url = {https://huggingface.co/harleensachdev/yolo-lanyard-object-detector}
}

License

MIT License - See LICENSE file for details

Acknowledgments

• YOLOv8: Ultralytics (https://github.com/ultralytics/ultralytics)
• Goal: SDG 16 - Peace, Justice and Strong Institutions
• Project: Goals in Code Initiative

Contact

For questions or collaboration: [Your contact information]

---

Last Updated: 2026-02-01