trellixvulnteam / har-data-collection_ckes

First Part of the Full-Stack Project

This git repo is basically Part 1 of the 3-Part full-stack data science project of Human Activity Recogntion. The entire project was designed with one particular use case in mind: an simple iOS App consuming phone sensor data (accelerometer, gyroscope, etc..) fetched from the phone itself and utilizing a trained ML model to make inference from those fetched data about the activity of the human being

Part 1: Collecting phone sensor data wirelessly with basic iOS App (built with Kivy) and Websocket Server

Part 2: Building and Fine Tuing Neural Net Model with Keras/TensorFlow and "Weight and Bias"

Part 3: Deploying Keras Model on iOS device (CoreML and TF Lite) (upcoming)

########################################################################################################################################################

PART 1: DATA COLLECTION

I have almost very little experience with Kivy and Websocket when I had this idea of collecting data wirelessly. This part of the project would not have been possible without the turtorial from: [https://www.youtube.com/watch?v=uv8PciALzSI]

A couple of lessons and modification I had achieved throghout this part:

• Learn about different communication prototcol of edge devices
• Change the sampling rate of the sensor data and include extra attribute/features on the client script to make it more similar to a public HAR dataset
• Change the server script to stop the streaming period after certain amount of time and write to a JSON file