Functional activity monitoring from wearable sensor data

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Monitoring unscripted, real-world physical activities presents significant challenges.
• Existing methods often struggle with the variability and overlap inherent in freeform human movement.
• Accurate interpretation of sensor data is crucial for understanding functional motor activities.

Purpose of the Study:

• To present a novel approach for interpreting EMG and accelerometer data.
• To enable monitoring, identification, and categorization of functional motor activities in unrestrained environments.
• To develop a practical system for real-world activity recognition.

Main Methods:

• A hierarchical approach combining artificial intelligence (AI) and neural networks.
• Utilizing blackboard and rule-based systems for coarse-to-fine activity partitioning.
• Employing integrated processing and understanding of signals (IPUS) for classification refinement.
• Leveraging neural networks for initial activity classification.

Main Results:

• The proposed method effectively partitions and classifies functional motor activities.
• It accounts for signal variability and overlap in unscripted movements.
• Provides a practical framework for real-world activity monitoring using sensor data.

Conclusions:

• The novel approach offers a practical solution for interpreting sensor data in real-world settings.
• This method enhances the accuracy of identifying and categorizing functional motor activities.
• It represents a significant advancement in human activity recognition technology.