README.md

---
language: en
license: mit
library_name: transformers
tags:
- computer-vision
- drowsiness-detection
- driver-safety
- cnn
- tensorflow
model_name: drowsiness_model.h5
datasets:
- ckcl/drowsiness_dataset
- custom
metrics:
- accuracy
- binary-crossentropy
widget:
- text: Example input
pipeline_tag: image-classification
base_model:
- google/mobilenet_v2_1.0_224
---

# Driver Drowsiness Detection Model

This model is designed to detect driver drowsiness from facial images using a CNN architecture.

## Model Details
- Architecture: CNN
- Input: Facial images (64x64x3)
- Output: Binary classification (drowsy/not drowsy)

## Usage
```python
import tensorflow as tf
import cv2
import numpy as np

# Load model
model = tf.keras.models.load_model('drowsiness_model.h5')

# Preprocess image
img = cv2.imread('face.jpg')
img = cv2.resize(img, (64, 64))
img = img / 255.0
img = np.expand_dims(img, axis=0)

# Make prediction
prediction = model.predict(img)
is_drowsy = prediction[0][0] > 0.5
```

## Training Details
- Dataset: Custom driver drowsiness dataset
- Training method: Binary cross-entropy loss with Adam optimizer
- Validation split: 20%
- Early stopping with patience=3

## Model Architecture
- Input Layer: 64x64x3 images
- Convolutional Layers:
  - Conv2D(32, 3x3) + BatchNorm + ReLU
  - MaxPooling2D(2x2)
  - Conv2D(64, 3x3) + BatchNorm + ReLU
  - MaxPooling2D(2x2)
  - Conv2D(128, 3x3) + BatchNorm + ReLU
  - MaxPooling2D(2x2)
- Dense Layers:
  - Dense(128) + BatchNorm + ReLU
  - Dropout(0.5)
  - Dense(1) + Sigmoid

## Performance
- Binary classification for drowsiness detection
- Optimized for real-time inference
- Suitable for embedded systems and edge devices

## License
This model is released under the MIT License.