A Hybrid Approach to Estimate Vertical Oscillation Using a Chest-Worn Accelerometer Device

• Published in: Medical Measurement and Applications (MEMEA), IEEE International Workshop on pp. 1 - 6
• Main Authors: Nair, Shrihari, R, Dhinesh, V, Sricharan, S P, Preejith, Sivaprakasam, Mohanasankar
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.06.2024
• Subjects: Accelerometers; Computational modeling; Feature extraction; inertial sensors; machine learning; motion capture; Prevention and mitigation; Real-time systems; Sensor phenomena and characterization; sports technology; vertical oscillation; Wearable devices; wearable technology
• ISSN: 2837-5882
• DOI: 10.1109/MeMeA60663.2024.10596708

Monitoring the vertical oscillation (VO) of an individual aids in improving gait mechanics, minimizing energy consumption, and lowering the chances of injury. Wearable devices equipped with Inertial Measurement Units (IMUs) offer real-time monitoring of VO, aiding in movement optimization and injury prevention. However, calculating VO in real-time on wearables consumes significant power due to the utilization of multiple sensors, thereby impacting the longevity of the device. Therefore, a potential power-efficient solution utilizing a single accelerometer and a lightweight algorithm to compute VO is developed and is compared with IMU-based motion capture systems for ground truth validation. A study involving 26 participants utilizes the chest-worn Netrin sensor, equipped with an accelerometer, to record vertical motion while they run on the treadmill. The collected data is then processed to compute the vertical displacement of the subjects. To mitigate errors arising from sensor drift and the absence of orientation information, a multiple linear regression model is applied, with the features extracted from the vertical motion signal serving as inputs to the model. The model predicts VO with a mean absolute error (MAE) of 0.65 cm and a mean squared error (MSE) of 0.75 cm 2 , with a coefficient of determination of 0.75. Successfully tested on field runners, this hybrid approach combining both traditional and machine learning methods, shows promise despite hardware constraints, enabling real-time VO computation with minimal errors via a straightforward data processing algorithm.