An Android App named PDIoT which implements real time human activity recognition via blue tooth connectable RESpeck sensor.
This repo contains:
HAR_pdiot_app: the APP itself scripted in Java and Kotlin.
raw_data: a public HAR dataset collected from 49 subjects
with RESpeck sensor. Each subject records an 30s data of each activity class, stored in Respeck_UUN_class_date_time.xlsx file.
classifier for 5 classes: a tutorial of applying CNN and GRU on the above dataset in Tensorflow. Learned tf-lite model will
be deployed on App to achieve online classification. The best model can get will have accuracy of 95% on testing set.
Open HAR_pdiot_app via Android Studio, open developer mode of target Android device, find the device in
Android Studio and press run.
Below shows the App interface. From left to right are main page, RESpeck sensor connection page
and prediction page. The App has a extra sedentary reminder functionality.
This small demo shows the App classification result when subject is falling.
