IMU Gesture detection with a Neural Network

Using an XinaBox CW01 (ESP8266) and XinaBox SI01 (LMS9DS1 9 DoF IMU)

This Project uses the CW01 and SI01 to collect a total of 1350 samples from the IMU in 1.5 seconds (split between the Magnetometor, Accelerometer, and Gyroscope)
This data is then sent via a POST request to a server which is run with predict.py, which reads in the data and makes a prediction using classifier.model,
which is a pre-trained neural network which can be downloaded here (131MB)

If you want to train the model to recognize different gestures, use getData to get the POST request and read the data to a csv, you can then use TrainGestures.ipynb to train the data.

Files

• SendData/SendData.ino
  • Reads and sends the 1350 samples to the server
• getData.py
  • Runs a server and appends the contents of the POST request to a csv
• predict.py
  • Runs a server and makes a prediction based on classifier.model
• TrainGestures.ipynb
  • Use to plot data from the csv, train and save the neural network model
• classifier.model
  • The saved trained neural network
  • Download here (131MB)

Libraries

• Arduino

  • ESP8266WiFi.h
  • ESP8266HTTPClient.h
  • xCore.h
  • xSI01.h

• Python

  • http.server
  • pickle
  • pandas
  • numpy
  • scipy
  • logging
  • matplotlib
  • seaborn
  • sklearn