AI based detection of tiredness

To work efficiently it is important to make pauses as soon as there are symptoms of tiredness. It is useful to have automatic reminders based on individual physical features. That's where the instruments of artificial intelligence can help.

• Physical parameters

To measure tiredness I used two factors: my height when I sit and the degree of eye openness. I followed the logic that the more tired a human becomes, the more relaxed his back will be, so the sitting height will decrease and the degree of eye openness will be decreasing to reduce the amount of absorbed light.

• Artificial intelligence model

So keeping these two factors in mind, I trained an online-model on Techable machine site and prepared two classes: the first class contained images with me sitting straight in front of the computer, my eyes widely opened. The second class contained images with me sitting in more relaxed positions, with my eyes being half-closed.

Then I checked if my model can work efficiently with changed parameters: different clothes, the presence of the glasses, modified haircut, changed background or light. The experiments showed that the model is reliable with the change of all parameters, except for the background's change. Here are the results:

So when there is a working model, it should be exported for further work:

• Scratch code for automatic measure with micro:bit

I used Vittascience Adacraft platform to prepare a code that will allow to check how tired I am and to show information about my state on micro:bit display. More precisely, every 60 seconds there is a check of a class determination via prepared Teachablemachine model, if for at least 15 minutes in a row the second class is determined (the class that shows the signs of tiredness), then the word "TIRED" will be displayed on micro:bit and the luminosity level will be shown; otherwise the word "GOOD" will be displayed and the luminosity level will be shown as well.

I attach a file with the code and the link.

• Test of the program

To try this project you need an AI model with your images, the code and connected micro:bit. The electronic board will display an information about the state of tiredness every minute.