• Phone: +1 844-COGNITRO
  • info@cognitro.com
Stay Connected:

Fatigue.Ai

Fatigue measurement has been undertaken using physiological measurements, but a non-invasive way of accurate detection of fatigue remains under investigation. Fatigue.AI is an image-based embedded solution for detecting drivers’ signs of fatigue and tendencies to fall asleep during driving in real-time. It works by capturing and analyzing critical features related to sleep and drowsiness over short but regular time intervals. These include micro-sleeps, eye-blinking, yawing opening, as well as head tilts and orientation. Using deep learning, the system has been trained to extract those features using a reference and well-established data sets of videos produced by subjects at various levels of fatigue tagged according to KSS (Karolinska Sleepiness Scale). A machine learning-model was then used to classify new instances of driving videos based the extracted parameters. The model can be run on the cloud but a new embedded model is under development on high-speed edge computing portable device that can be mounted on the car dashboard.

How Does It Work


1) FACE RECOGNITION
The driver take the driving seat with a mounted web-cam facing the drivers head. Driver’s face and contact information is authenticated from the system registry in order to alert the driver or an authorized party in case of an elevated level of fatigue.
2) LANDMARKS DETECTION
The system localizes the critical regions related to fatigue including the eye, mouth and face contour. A set of model parameters are extracted including PERCLOS (percentage of eye close) , micro-sleeps, eye blink duration, yawning, etc.
3) FATIGUE DETECTION
The extracted parameters are passed to a pre-trained machine learning. The detected fatigue indicator displays the fatigue state of the driver on a KSS scale expressing alertness state vs. fatigue level.

Fatigue.Ai in Action