Classifying movements on flooring based on machine learning and capacitive sensing


In an aging society, safety and care of elderly people are increasingly receiving attention. New technologies are gaining importance in supporting care services, in simplifying life, and in controlling smart homes. Determining human presence and classifying human activities are two pillars of advanced control in smart homes that build upon strategies for detecting people. However, current strategies rely on optical sensors, which need light and a direct visual contact to reliably detect persons. Therefore, the application of optical sensors is limited and collecting sensor data may violate the personal privacy. Systems without optical sensors, capable of determining more than just the presence of people may help detect falls, analyze walking routes, and check distances between people in pandemics without violating personal rights.

Project goals

This project aims to develop a methodology for leveraging the data collected by capacitive sensors in floorings to differentiate human activities, including, e.g., a differentiation between humans and pets. A main objective is to investigate the information entropy in an attempt to generate a maximum of information from a single sensor to minimize the total number of sensors, thus keeping the data volume small and possible use cases economically feasible. The main component of a sensor will be a conductive sensor surface (Figure 1) measuring approximations (< 30 cm) as well as direct contact to the floor. In addition, a machine learning (ML) algorithm will be designed and trained on the variety of possible footsteps, able to analyze (i) single and (ii) combined sensor data.

Expected outcome and societal impact

The methodology to be developed in this project may be used in a variety of applications, e.g. for security systems, for fall detection in medical institutions, or for controllers in smart homes. In public buildings and exhibition halls, the methodology may deliver statistics or support checking distances between human individuals in pandemics. As capacitive sensors not only react to changes in distance to an object but also to humidity, object recognition and safety systems in collaborative robotics are further applications to be advanced by the methodology proposed herein.

Project type

PhD project in cooperation with Bern University of Applied Sciences.


Professor Dr. Kay Smarsly
Hamburg University of Technology
Institute of Digital and Autonomous Construction
Blohmstraße 15
21079 Hamburg
Email: kay.smarsly(at)tuhh(dot)de

Professor Dr. Thomas Volkmer
Bern University of Applied Sciences
Solothurnstrasse 102
2500 Biel
Email: thomas.volkmer(at)bfh(dot)ch

Filipp Wirth, M.Sc.
Bern University of Applied Sciences
Solothurnstrasse 102
2500 Biel
Email: filipp.wirth(at)bfh(dot)ch