direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Page Content

Human Activity Recognition using Smartphone and Smartwatch Sensors

In many areas of networking, including IoT systems, the question of data and processing task placement arises. Potential candidates for storage and processing would be phones, gateways, and fog or Cloud virtual machines. Finding the optimal placement strategy requires knowledge about the future usage pattern and context information, such as user connectivity (Wi-Fi, 3G, intermittent connectivity). Nowadays, modern off-the-shelf smartphones and smartwatches have an increasingly rich set of embedded sensors, such as accelerometer, gyroscope, compass, WiFi, and GPS. This enables to capture the contextual information needed from the underlying human behavior in real-time and predict future behavior and data needs.
The goal of this project is to asses the feasibility of human activity recognition using smartphones and smartwatches in the specific IoT context. Students are expected to apply various state-of-the-art machine learning algorithms to detect and predict human behavior to improve data placement. Students should further evaluate the trade-off between accuracy of the solution and limiting factors of the hardware, such as impact on battery life or mobile data usage. The project will roughly follow the following steps:

  • Enable logging of phone and watch sensors on Android
  • Measurement campaign of different activities that are to be recognized
  • Adaptation of machine learning frameworks for activity recognition
  • Evaluation of activity recognition (accuracy, battery, etc.) in the context of data and processing task placement.
  • Advanced programming skills in Java
  • Basic understanding of Android
  • Deep understanding of machine learning concepts
Daniel Happ ()




Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions