Drowsiness and Distraction

Distraction Methodology

1. The algorithm starts after face recognition is finished so it will not start at ignition.

2. The first step for distraction is to establish the normal position of the driver, so it collects every 3 seconds the position of your face key landmarks points to establish a baseline (normal driver position). After collecting 10 samples, it is starting the detection procedure. This does not mean that the detection will start after exactly 30 seconds after FR in finished, the algorithm sometimes cannot detect face landmarks and it takes more time to gather 10 samples.

3. Distraction algorithm is continuously extracting key landmarks and does 2 things:

a) comparing the current samples with the mean of the baseline. If the difference exceeds a certain threshold in lower Y axis transition and driver head orientation on Y axis (if your face key points are way down and head orientation is also oriented down compared to driver normal position) that will give a trigger

b) updating the baseline with new key points position detection and removing the oldest detection (dynamic baseline)

4. Distance from the driver to the camera can influence the algorithm. if you are standing closer to the camera and looking down you are moving a larger distance in pixels comparative to when you are further. Our algorithm is auto calibrating for any distance using the distance between your eyes as a measurement of how far you are from the camera and making the threshold dynamically. But if you stay too far away from the camera, the algorithm could not detect your face properly.

Drowsiness Methodology

1. The algorithm starts after face recognition is finished so it will not start at ignition.

2. Drowsiness does not need a baseline, so it will start detection exactly after face recognition ends.

3. Drowsiness algorithm is continuously extracting key landmarks and calculating EAR (eye aspect ratio - height of eye/ length of the eye). If this report is under a certain threshold for more than a number of seconds the algorithm will trigger.

In field functionality and limitations

The camera should be placed either right or left of the driver!!! The feature will not work if the camera is place exactly in front of the drivers face This is an algorithm designed for the hardware possibilities of our camera. So, do not expect to work perfectly. Due to hardware, illumination conditions, camera position limitations the algorithm is limited and many times susceptible to errors. Sometimes the illumination conditions don’t allow the detection of drivers face or key points in real-time. So, it will miss triggers Other times the fps does not catch the transition of looking down at a cell phone or other devices. Other times when the driver is readjusting his place in the seat and getting in a much lower position can trigger. Also, might be errors due to driver looking in the mirror and moving the head down while doing that Sun glasses and normal glasses can influence face key points detection and drowsiness algorithm. (camera cannot see the eyes while the driver is wearing sun glasses and eyes dimensions ca be modified by normal glasses lens) Most of the real distraction events are happening while the car is not moving (standing on light stops, etc) so do not expect a lot of events while minimum speed is set to a value larger than 0.

Parameters that can be custom modified for customers

• Minimum speed from which this feature is starting to be active and can give triggers (default is 5km/h)
• How much time does a driver needs to stay with his eyes closed in order to trigger (default is 2.5 seconds)