The application of EMD in activity recognition based on a single triaxial accelerometer

Area of Science:

• Wearable technology
• Biomedical engineering
• Signal processing

Background:

• Activity recognition using wearable sensors is a key research area.
• Accelerometers are versatile and widely used sensors for this purpose.
• Existing methods may face limitations due to sensor placement and user constraints.

Purpose of the Study:

• To propose a novel method for activity recognition using a single accelerometer.
• To extract relevant features from accelerometer data using statistical methods and empirical mode decomposition (EMD).
• To evaluate the performance of the proposed method across different sensor placements.

Main Methods:

• Feature extraction using statistical measures (mean, standard deviation, entropy, energy, correlation).
• Advanced feature extraction via empirical mode decomposition (EMD) to obtain intrinsic mode functions (IMFs).
• Classification of activities using the Adaboost algorithm.

Main Results:

• High classification accuracies achieved: 94.69% (waist), 86.53% (left thigh), 91.84% (right ankle), and 92.65% (right arm).
• The method demonstrates effectiveness regardless of sensor position.
• The approach mitigates issues related to movement constraints and user discomfort.

Conclusions:

• The proposed single-accelerometer method offers a robust and accurate solution for activity recognition.
• EMD-based feature extraction enhances performance for non-linear and non-stationary accelerometer data.
• The flexibility in sensor placement broadens the applicability of wearable activity recognition systems.