The SensorTemplate provides a simple build of the Android application with centrelized tools to manipulate with sensors of a phone. Whole computational unit is build on a foreground service, which can be turn on by the button in the main activity.
The model under foreground service connects different aspects of Android system together.
- whole life cycle of sensors is handled
- temporal storage for samples
- creation of csv files, if needed
- implementation of GPS
- implementation of Activity Recognition Client
- methods to handle changes in network, user activity and battery life
- wakelock handling
- app is added to whitelist for background processing by AutoStarter
- examples of detector classes
Every aspect is completely modifiable.
Tensorflow lite neural network for 1D signal/features. Model can be also modified to 2D if needed.
Scikit-learn models like SVM, decision trees, k-neighbours classifier, Naive Bayes has been added to app with help of sklearn-porter. Outputs have been modified into the Kotlin, with one interface. Parameters for models should be extracted with help of sklearn-porter.