DTWSpells-AshwinAkshit

Harry Potter Spells 🧙🏼‍ 🎩

Ashwin Sankaralingam, Akshit Arora

---

Sensors used : Two microbits

• Wand Microbit :
  • Send accelerometer reading via bluetooth.
  • Display the icon on LED based on received signals.
• Laptop Microbit :
  • Send received signals via serial port to Wekinator (pyscript).
  • Send classified prediction (spell) to display in wand.

Feature Extraction:

Just stream accelerometer readings speedup using Ben's hack.

Feature Engineering: Using totally varied gestures (❤️,🐍,💀)

ML block:

We used DTW for training our examples over three classes. In order to avoid confusion, all the three gestures have a varied start point, involves different motion and direction.

Experiments:

We tested out two versions of DTW:

1. Matches computed only after running stops : Even though this method did really well, it was painful to stop and start the running everytime at the end of the spell.
2. Matches computed continuously while running: This started to working better when we started to play with the thresholds and find optimal values through stochastic testing.

Improvements we could have made:

1. Insted of sending x,y,z (accelerometer) reading continuously, we could have done something like dx,dy,dz and found some interesting patterns.

Github link : https://github.com/CUBoulder-2019Sp-IML4HCI/DTWSpells-AshwinAkshit

Youtube link : https://www.youtube.com/watch?v=sT3Nnmz4WO4&feature=youtu.be