ioanachelaru / drowsiness-detection-system

Drowsiness detection with OpenCV and CNN

The poject is build to detect early signs of fatigue for drivers based on their eye movement using a real-time video stream and playing an alarm when the divers appears to be drowsy.

Dependencies

The project uses cv2, keras and tensoflow. All the required dependencies can be found in requirements.txt.

Algorithm

The face is detected using haar-like features since is a quick and efficient enough method and the eyes are detected using dlib's shape_predictor_68_face_landmarks.dat because is a more effective method than the previous one.

If both eyes are detected they are passed to a convolutional neural network build and trained for their classification.

If both eyes are classified as closed, a counter will increase its value, otherwise, it will be decreased. When the counter reaches a threshold, the driver will be declared drowsy, an alarm will start ringing, and a red border will be added to the application's window. The alarm will stop and the border will disappear when the value of the counter will be again smaller than the threshold.

The Convolutional neural network

It was built and trained using Google Colaboratory and a dataset of around 3000 images of open and closed left eyes.

The CNN has been trained with only left eyes, that way creating a binary classifier since we're not interested in differentiating the right from the left eye. For that reason, before sending the segment containing the right eye detected, a flip operation is applied over it so it simulates the left eye distribution.