Drowsiness Detection
Introduction
This project is aimed towards solving the problem of detecting the tiredness level of drivers. If the driver is found to be drowsy, an alarm will be triggered to make him or her more attentive. This is done using the techniques like deep learning and computer vision for object detection. The project was implemented on a raspberry pi and a video camera.
Hardware Required:
- Raspberry Pi 3
- Picamera
This project uses python 3 along with openCV for image processing and dlib for facial landmarks detection. Since raspberry pi provides limited memory and processing speed, we used a combination of Haar Cascades and HOG to detect the face and the eyes of the driver.
Haar cascade was used for detecting the face since it is inexpensive from a computation standpoint whereas HOG was used for detecting eyes in the face. HOG although, computationally expensive provides accurate boundaries for eyes. To perform this successfully, we used dlib, which is a free to use library on python to detect eyes using HOG.
Once the boundary of the eyes are found, we then find the aspect ratio, ie length/height of the eye and if it is below a certain threshold the program triggers an alarm
Setup
Note:The user should be comfortable using linux based operating systems and raspberry pi.
Install OpenCV, Dlib and Numpy on your raspberry pi. Download the repository and connect the picamera module if you are using a USB web camera, you need to make a slight change in the code. on the line no. 80, set the usePicamera=False. You also need to download the .dat file for dlib facial landmarks.
Run the following command in the program directory:
python drowsiness.py --cascade haarcascade_frontalface_default.xml --shape-predictor shape_predictor_68_face_landmarks.dat