Online Activity Recognition Combining Dynamic Segmentation and Emergent Modeling

Area of Science:

• Pervasive Computing
• Human-Computer Interaction
• Artificial Intelligence

Background:

• Activity recognition is crucial for smart environments and pervasive computing applications.
• Existing methods often rely on pre-segmented data, limiting real-time application.
• Real-world deployments necessitate online activity recognition from continuous sensor streams.

Purpose of the Study:

• To propose an online activity recognition model for real-time analysis of streaming sensor data.
• To address the limitations of traditional activity recognition methods in dynamic environments.
• To enable accurate mapping of sensor data to human activities as they occur.

Main Methods:

• A dynamic segmentation approach using spatio-temporal correlations to define event windows.
• An emergent modeling method based on stigmergy to build activity features.
• Representation of activity features as a directed weighted network for context definition.

Main Results:

• The proposed model effectively recognizes activities from streaming sensor data in real time.
• Validation using the Aruba dataset from the CASAS project demonstrates the method's effectiveness.
• The approach successfully segments sensor events and models emergent activity patterns.

Conclusions:

• The developed online activity recognition model is effective for smart environments.
• The combination of dynamic segmentation and stigmergy-based modeling offers a robust solution.
• This method provides a valuable tool for real-time human activity understanding in pervasive computing.