Person-Specific Template Matching Using a Dynamic Time Warping Step-Count Algorithm for Multiple Walking Activities.
Valeria FilippouMichael R BackhouseAnthony C RedmondDavid Chi-Wai WongPublished in: Sensors (Basel, Switzerland) (2023)
This study aimed to develop and evaluate a new step-count algorithm, StepMatchDTWBA, for the accurate measurement of physical activity using wearable devices in both healthy and pathological populations. We conducted a study with 30 healthy volunteers wearing a wrist-worn MOX accelerometer (Maastricht Instruments, NL). The StepMatchDTWBA algorithm used dynamic time warping (DTW) barycentre averaging to create personalised templates for representative steps, accounting for individual walking variations. DTW was then used to measure the similarity between the template and accelerometer epoch. The StepMatchDTWBA algorithm had an average root-mean-square error of 2 steps for healthy gaits and 12 steps for simulated pathological gaits over a distance of about 10 m (GAITRite walkway) and one flight of stairs. It outperformed benchmark algorithms for the simulated pathological population, showcasing the potential for improved accuracy in personalised step counting for pathological populations. The StepMatchDTWBA algorithm represents a significant advancement in accurate step counting for both healthy and pathological populations. This development holds promise for creating more precise and personalised activity monitoring systems, benefiting various health and wellness applications.
Keyphrases
- machine learning
- deep learning
- physical activity
- neural network
- artificial intelligence
- big data
- healthcare
- public health
- mental health
- peripheral blood
- human health
- mass spectrometry
- cross sectional
- depressive symptoms
- lower limb
- patient reported outcomes
- blood pressure
- risk assessment
- molecularly imprinted
- heart rate
- social media
- solid phase extraction