Paper Info
Title
Wavelet-based detection of gait events from inertial sensors: analysis of sensitivity to scale choice
Abstract
Nowadays, inertial sensors have shown great popularity in gait analysis, thanks to their advantages in terms of costs, encumbrance and usability in outdoor settings. The accuracy achieved by inertial-based systems in detecting gait events is however not as high as that obtained with lab-based equipment, even if many authors have presented methods to fill this gap. In particular, the segmentation of gait patterns into different cycles, and the determination of the stance and swing parts of each cycle are two key phases in the gait analysis process, as they are needed to extract all the spatio-temporal parameters. Thus, an accurate detection of initial contact (IC) and final contact (FC) events is necessary. Among the different methods presented in the literature, we chose a wavelet-based one, which used the vertical component of the acceleration extracted from an inertial sensor on the waist, and we investigated the changes in detection accuracy when varying the numerical values of a parameter (i.e., the wavelet scale value). Gait patterns of five healthy subjects were recorded by a waist-mounted inertial sensor, and by a stereophotogrammetric system, taken as reference: IC and FC events were then extracted and compared, obtaining an estimation for each scale. Precision, Sensitivity, and $\rm F1_{\mathbf {score}}$ curves showed that IC events are well estimated in the range of scales 29–87, while FC events in the range 24–65, at a sampling frequency of 250 samples/s. In these ranges, mean error, variability and RMSE values showed lower values for the estimated IC events (RMSE<39 ms) than for the FC ones (RMSE<47 ms). A common accurate subrange of scales was then determined (29–37). Thus, this analysis confirmed the good accuracy of the wavelet-based method within a wide range of scales and showed how the scale choice influences the accuracy of the estimates.
Year
DOI
Venue
2018
10.1109/MeMeA.2018.8438651
2018 IEEE International Symposium on Medical Measurements and Applications (MeMeA)
Keywords
Field
DocType
motion analysis,gait events,inertial measurement units,wavelet
Pattern recognition,Gait,Sampling (signal processing),Mean squared error,Gait analysis,Inertial measurement unit,Artificial intelligence,Motion analysis,Mathematics,Instrumentation,Wavelet
Conference
ISBN
Citations 
PageRank 
978-1-5386-3393-9
0
0.34
References 
Authors
4
6
Name
Order
Citations
PageRank
Carlotta Caramia151.22
Ivan Bernabucci2144.10
Carmen D'Anna302.70
Cristiano De Marchis4125.89
Andrea Scorza524.61
Maurizio Schmid68916.32