CFE 2019: Start Registration
View Submission - CMStatistics
B1889
Title: Clustering time-varying quaternion data: Application to the detection of gait abnormalities Authors:  Pierre Drouin - Laboratoire de mathematiques Jean Leray - Nantes / UmanIT (France) [presenting]
Aymeric Stamm - CNRS (French National Center for Scientific Research) (France)
Lise Bellanger - Nantes University (France)
Laurent Chevreuil - UmanIT (France)
Vincent Graillot - UmanIT (France)
Abstract: Wearable motion sensors have become more and more compact and affordable. As such, they might become useful to the neurologist for achieving objective and quantitative assessment of walking disability in patients diagnosed with a neurodegenerative disorder, such as Multiple Sclerosis. The company UmanIT and the Department of Mathematics Jean Leray in Nantes have developped a solution for facilitating gait analysis by collecting data using an inertial measurement unit embedded in a motion sensor. The device is clipped on the belt and records hip rotation over time as a sequence of quaternions. After pre-processing, a set of walking cycles (i.e. all movements made between two equivalent positions of a given foot) is provided by the device. To test the hypothesis that differences between hip motion during walking cycles are related to walking disability, an experiment was conducted on healthy volunteers who performed a walking test under two conditions: (i) free motion and (ii) wearing a knee blocking splint. Clustering methods will be adapted to account for the non-Euclidean nature of quaternion data. In particular, approaches borrowed from time series (e.g. dynamic time warping) and from functional data (e.g. k-mean alignment) will be extended. As part of a benchmarking study, these methods will be evaluated based on their performance to create two groups that separate walking cycles according to the two test conditions and compared to traditional clustering methods.