A method to deal with installation errors of wearable accelerometers for human activity recognition

Area of Science:

• Biomedical Engineering
• Wearable Technology
• Human Activity Recognition

Background:

• Human activity recognition (HAR) using wearable accelerometers is crucial for healthcare applications like fall prediction and gait analysis.
• Accurate HAR depends on correct sensor placement; installation errors significantly reduce system accuracy.
• Existing HAR methods are sensitive to sensor orientation and placement inaccuracies.

Purpose of the Study:

• To develop a robust method for human activity recognition (HAR) that mitigates the impact of wearable accelerometer installation errors.
• To improve the reliability and accuracy of HAR systems in real-world healthcare scenarios.
• To address the challenge of sensor misplacement and orientation errors in wearable accelerometer data.

Main Methods:

• A transformation matrix is calculated using Gram-Schmidt orthonormalization to correct sensor orientation errors.
• A low-pass filter with a 10 Hz cut-off frequency is applied to minimize the effects of sensor misplacement.
• The proposed method integrates orientation correction and filtering to enhance HAR robustness.

Main Results:

• The proposed method demonstrated satisfactory performance in human activity recognition (HAR).
• Accuracy remained high at 91.9% even with intentional installation errors in wearable accelerometers.
• Without installation errors, the average accuracy rate across ten subjects was 95.1%.

Conclusions:

• The developed method significantly improves the robustness of HAR systems to wearable accelerometer installation errors.
• This approach enhances the reliability of HAR for diverse healthcare applications.
• The technique offers a practical solution for maintaining HAR accuracy despite sensor placement variability.