That’s one way to go. But not so good. A zig-zag-line is just YOUR understanding of angry.
Instead, when 10 people played your game, you have 10 images for each emotion. For a new image, the AI (openCV) checks which 10 images of each emotion are in average most similar to the new one (the emotion with the highest average score wins). I tried to describe this above. So you don’t rely on rules (zig-zag-line) but on the gathered images (more statistically).
The term AI
AI is a broad term though. OpenCV would not really qualify as AI. A neural network would. A neural network would learn (be trained) from your 10 example images for each emotion and then give answers. There must be libraries for this.
Check out Neural Network
tutorial https://www.youtube.com/watch?v=XJ7HLz9VYz0
check out the forum search Search results for 'neural' - Processing Foundation
for example Improving the accuracy of a neural network for digit recognition - #2 by SomeOne
Chrisir