nnsohamnn/helmet-detection-yolo11

base_model: - Ultralytics/YOLO11

🪖 Helmet Detection using YOLO11

This model detects motorcycle riders wearing helmets and those without helmets.

Classes:

• With Helmet
• Without Helmet

Framework: Ultralytics YOLO11
Task: Object Detection

---

🧠 Helmet Violation Detection — System Flowchart

flowchart TD
    A[📥 Input Image / Video Frame] --> B[🔍 YOLO Scene Detection]

    B --> C1[🧍 Detect Persons]
    B --> C2[🏍️ Detect Motorcycles]

    C1 --> D[📐 Rider Matching Logic]
    C2 --> D

    D -->|Person overlaps Motorcycle| E[🏍️ Identified as Rider]
    D -->|No overlap| F[🚶 Ignore (Pedestrian)]

    E --> G[✂️ Crop Rider Head/Upper Body]
    G --> H[🪖 Helmet Detection Model]

    H -->|Helmet Detected| I[🟢 Safe Rider]
    H -->|No Helmet Detected| J[🔴 Helmet Violation]

    I --> K[🖼️ Draw Green Box + Label]
    J --> L[🖼️ Draw Red Box + Label]

    K --> M[📤 Output Image with Results]
    L --> M

---

🚀 Quick Inference in Google Colab (Single Cell)

Copy and run the entire cell below in Google Colab:

# ================== HELMET DETECTION — HF MODEL (ONE CELL) ==================

!pip install ultralytics -q

from ultralytics import YOLO
from google.colab import files
import cv2
import matplotlib.pyplot as plt

# 🔹 Hugging Face model link (URL-encoded)
model_url = "https://huggingface.co/nnsohamnn/helmet-detection-yolo11/resolve/main/yolov11m%28100epochs%29.pt"

print("📥 Loading model from Hugging Face...")
model = YOLO(model_url)
print("✅ Model loaded!")

print("\n📤 Upload images for helmet detection...")
uploaded = files.upload()

for filename in uploaded.keys():
    print(f"\n🖼️ Processing: {filename}")

    img = cv2.imread(filename)
    if img is None:
        print(f"❌ Could not read image: {filename}")
        continue

    results = model(img, conf=0.35)[0]
    img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

    for box in results.boxes:
        x1, y1, x2, y2 = map(int, box.xyxy[0])
        cls = int(box.cls[0])
        conf = float(box.conf[0])
        label = model.names[cls]

        # Color coding
        if "without" in label.lower():
            color = (255, 0, 0)  # 🔴 No Helmet
        else:
            color = (0, 255, 0)  # 🟢 Helmet

        # Draw bounding box
        cv2.rectangle(img_rgb, (x1, y1), (x2, y2), color, 3)

        # Draw label background
        text = f"{label} {conf:.2f}"
        (w, h), _ = cv2.getTextSize(text, cv2.FONT_HERSHEY_SIMPLEX, 0.6, 2)
        y_text = y1 - 10 if y1 - 10 > 10 else y1 + h + 10
        cv2.rectangle(img_rgb, (x1, y_text - h - 4), (x1 + w, y_text), color, -1)

        # Put label text
        cv2.putText(img_rgb, text, (x1, y_text - 2),
                    cv2.FONT_HERSHEY_SIMPLEX, 0.6, (0, 0, 0), 2)

    plt.figure(figsize=(8, 8))
    plt.imshow(img_rgb)
    plt.axis("off")
    plt.show()

print("\n✅ Detection complete!")
# ============================================================================

---

📈 Validation Metrics

Class	Images	Instances	Precision (P)	Recall (R)	mAP@50	mAP@50-95
All Classes	126	299	0.875	0.850	0.899	0.437
With Helmet	88	184	0.891	0.902	0.932	0.472
Without Helmet	53	115	0.860	0.798	0.866	0.402

---

🔍 Metric Meaning

• Precision: How many predicted helmets/no-helmets were correct
• Recall: How many actual helmets/no-helmets were detected
• mAP@50: Detection accuracy at IoU 0.50
• mAP@50-95: Stricter overall detection quality

📌 Summary

The model performs very well for detecting riders wearing helmets and shows moderate performance for detecting riders without helmets, which is typically harder due to occlusion and pose variation.