This Repository contains four applications to genereate training data for an IoT Machine Learning case to classify movements based on smartphone sensor data. It contains a node.js express frontend ("iot-smartphone-app") that can be accessed on the Browser with Android and iOS Smartphones. Sensor data of the accelerometer and gyroscope gets published/sent via secure Websocket over MQTT to the mosquitto MQTT Broker.

A python backend Service is subscribed ("Iot-subscriber") to the broker and writes everything to the database

local deployment

1. create pg-datenbank.env file and ibm-wml.env file and put it database and wml credentials; also adjust mqtt BROKER_URL to your local ip address in the docker-compose file

(optional) mosquitto TLS

use certs folder to generate certs with the script and provide ip address of your network interface in the Script "generate.sh" and name all server certs (those carry your ip/name) to server.*** set up mosquitto import cert to chrome (for web testing the connection to mqtt from your pc/mac) import cert to android (just send the ca.crt file via e.g whatsapp and install it)

2. docker-compose build

3. docker-compose up

local https serving (required because sensor works only in https context):

npx localtunnel --port 3005

remote deployment

mosquitto

create loadbalancer service get cert from ibm cloud cert manager put the certs into mosquitto deploy

frontend

set var for using local or remote mosquitto

backend

Notes

Interestingly devicemotion behaves differently on iOS and Android. on iOS it fires regularly on Android only on movement... that makes it a little difficult to compare. Neue "Generic Sensor API" for web wird auch nicht unterstützt. Es läuft wohl auf 2 Modelle hinaus....