ActiveSense – Sensor-Based Activity Recognition for Health and Care

Movement and activity data can help identify changes at an early stage, analyze care processes more objectively and, in the future, support documentation-intensive tasks. To achieve this, body-worn sensors, reliable data transmission and artificial intelligence need to work seamlessly together. ActiveSense combines these components in a modular research platform developed by Fraunhofer IGD.

How can body-worn sensor data be reliably captured, transmitted and translated into useful information for health and care?

From Sensor Data to Usable Information

ActiveSense captures movement and contextual data using a compact, body-worn sensor. The measurement data is synchronized over time, transmitted and stored on a data platform. From there, it is available for visualization, signal processing and AI-based analysis.

The modular architecture makes it possible to adapt sensors, communication channels and analytical methods to different research and application scenarios. The focus is therefore not on a single end device, but on the complete processing chain – from data acquisition on the body to interpretable information for research, healthcare and nursing care.

Robust Data Communication in Real-World Environments

For use in healthcare facilities, sensor data acquisition must remain reliable even where wireless connections are not consistently stable. The extended ActiveSense communication concept therefore incorporates local nodes and a self-organizing mesh network. Data can be forwarded across multiple nodes to the backend.

If a connection is interrupted, measurement data can be buffered locally and synchronized once the connection has been restored. Backend interfaces, the database and monitoring are designed as separate components. This makes it easier to add further sensors, dashboards, alerts or specific interfaces to existing applications.

AI-Assisted Recognition of Movements and Activities

ActiveSense processes multichannel time-series data from wearable sensors. After synchronization, noise reduction and segmentation, features or data-driven representations are generated and classified using machine learning methods.

The approaches investigated include Long Short-Term Memory networks (LSTMs) and Transformer architectures, which can model temporal patterns and dependencies in sensor data.

Controlled research studies have examined different movement and gait patterns, including normal and slow walking, shuffling steps, unstable gait patterns and asymmetric movements. In addition, the platform is being further developed to recognize other everyday and care-related activities.

The results provide a technical basis for future applications such as documentation support, process analysis, movement monitoring or preventive assistance systems. ActiveSense is a research platform and does not provide medical diagnoses.

Fraunhofer IGD’s Contribution

Fraunhofer IGD develops and integrates the core components of the ActiveSense platform. These include in particular:

• Design and integration of wearable sensors and sensor-level data acquisition
• Communication and backend architectures for continuous sensor data streams
• Data storage, visualization and monitoring
• Time-series analysis, feature extraction and machine learning pipelines
• Development and evaluation of LSTM, Transformer and other classification methods
• Application of the technical components to research questions in healthcare, nursing care, rehabilitation and motion analysis