Classification of physical activities based on body-segments coordination.

Area of Science:

• Biomedical Engineering
• Human Movement Science
• Wearable Technology

Background:

• Accurate physical activity (PA) recognition is crucial for health monitoring and rehabilitation.
• Existing methods often lack robustness and require sophisticated algorithms.
• Connected objects and PA monitoring innovations necessitate improved recognition methodologies.

Purpose of the Study:

• To develop an innovative and robust algorithm for physical activity recognition.
• To identify the optimal sensor data (accelerations) for distinguishing between different PAs based on body segment coordination.
• To validate the algorithm's performance across a diverse population.

Main Methods:

• Utilized heuristic definition of postures and body-segment coordination via external sensors.
• Computed vertical and horizontal accelerations from 8 anatomical landmarks using 3D motion capture.
• Compared 680 acceleration combinations using maximal Hausdorff Distance to find the best discriminators.
• Implemented a proof-of-concept algorithm using subject-specific reference data.

Main Results:

• Vertical accelerations from both knees proved most effective for discriminating between walking, running, and cycling.
• The proposed algorithm achieved 94% correct classification of physical activities.
• The methodology demonstrated validity for heterogeneous subjects, including varying ages and physical conditions.

Conclusions:

• A flexible and extendable methodology for physical activity recognition was proposed.
• The algorithm's effectiveness with diverse populations suggests potential for clinical and health applications.
• The approach leverages body segment coordination for robust and personalized PA monitoring.