Mood Cube

Mood Pillow

In this project, we were asked to create a "mood pillow" product that has at least three different "moods" (positions/sides/faces, etc) and then train our Teachable Machine to recognize these moods and display the current mood on our Arduino OLED display. Later, we were told to create a 2 minute video showcasing how it works as an advertisement for your product.

We decided to create a Mood Cube that displays 3 emotion in emoji faces: Happy, Sad and Angry. The users are able to orient the box according to how they feel so that other people will understand. It serves as a tool for communicating and conveying human emotions to the outside world, and is especially useful for people who have a trouble doing that. We also added LED lights in the place of the circular eyes of each emoji face. The lights turn on depending on the displayed face, in different colors: Red lights for the angry emoji face, blue lights for the sad and green lights for the happy one.

Onshape Drawing & 3D-Print

We created a rectangular box on Onshape that would fit the Arduino board inside, with an open top. After sketching 3 emoji faces for angry, sad and happy on each side, we used the Extrude command to remove the solid to get holes. For the fourth side, we further added a rectangular hole to fit the cable coming out of the Arduino board that would connect to the computer.

Later on, we added a lid at the top that read "Mood Cube" and includes a rectangular hole that would fit the display. Due to 3-D printing queue and NOLOP working hours, we could get it on the last day.

After 3D-printing our design, we placed the Arduino board and 6 LED lights in different pins. Pairing the LEDs with one another, we assigned each pair to one face and further taped them in the place of the eye holes. Later we adjusted the colors of the LED pairs accordingly using the screwdriver. We placed the screen at the side, at the place of the rectangular hole.

Teachable Machine

We created 3 classes for 3 different emotions: Happy, Sad and Angry. Using the webcam of the computer, we recorded image samples for each face, orienting the cube in different angles when introducing each face. At first, we got about 30 image samples for each face, however when we trained the machine and tested different faces on camera in Preview, we encountered some errors, like Sad face being detected as a Happy face. Then, we added more image samples, zooming out and in, and again, from many different angles to "teach better". When we tested again, the accuracy was slightly higher but still there were minor errors. We realized that the Angry emotion always showed higher accuracy when detecting because of its distinct expression with the brows, while Happy and Sad often did not have 90-100% accuracy for the output. Adding more image samples, we also increased the degree of "Epochs" this time since it says the larger the number the better the model will learn to predict data.

The screenshots of the Teachable Machine platform are seen below as we tested each side with a different face

Python & LabVIEW & SystemLink

When the front camera opened, we took 3 pictures of the 3 faces of the box, changing directions. Going to the TeachableMachines VI on LabVIEW, we introduced our converted_keras file and uploaded our pictures one by one. Regarding SystemLink, we created a String type tag called "emotion" and introduced it for SystemLink Control on the front panel. By this way, as the code detects the image through the Teachable Machine model, the result in higher percentage goes into SystemLink function and the function eventually writes it in a String format, which becomes the tag value. Each time we introduce a new picture, the tag value switches depending on the detected expression, Angry, Sad or Happy.

The testing of each picture (screenshot) on the Front Panel are displayed below, when we ran the ode and hit "Ask"

Then, we created if-else statements to meet 3 conditions in relation with 3 different expressions. When the tag value received from SystemLink is equal to one expression, the LEDs associated with that expression on the mood cube will light up. One of the issues was not writing the expressions for tag values in quotation marks at first, so it did not work. Then we were able to make the code work by adding the quotation marks. We identified each LED light pairs in relevant digital pins so that they would turn on at the same time as they belong to the eyes of the emoji faces. We further had to include other LEDs that belong to undisplayed expressions and set them to HIGH, meaning the lights should be off. By this way, when we switch emoji faces, as the tag value changes, the LED lights belonging to the previous expression would turn off and only the relevant LEDs will turn on.

The working process of our product, Mood Cube, while switching between different expressions is displayed below, through LabVIEW and Arduino.

Photographs of each side with emotions:

Movie

Thank you!