An Android App named PDloT which implements real time human activity recognition via bluetooth-connectable RESpeck sensor. This is a part of Principles and Design of IoT Systems course I took in 2022.
Below is how the app is designed. It can connect RESpeck sensor via bluetooth, and scanning QR code on sensor. Live Processing page showing live data from the sensor and activity classification result when a user is wearing RESpeck sensor. The machine learning model (CNN) deployed in app can achieve 18 activities classification, including user falling and direction of user fall. The classifier can achieve 96% accuracy on 5 activity classifications (AC) and 80% on 18 ACs. It prompts a sedentary reminder when a contious sitting if more than 1 hour is detected.
You can find project code in GitHub repo.
This demo shows classification results when a subject (me) is falling; it has a 2-second delay since we feed data sent from respect sensor via bluetooth in 2-second to the deployed in app TFlite model to get the result.