Supervised vs.unsupervised walking extraction – How does the recognition gets effected by sensor displacement?
Wearable sensor technology, particularly inertial measurement sensors are used in various motion analysis including applications in sport, medicine, and rehabilitation of patients after stroke. Wearability and sensor placement are related and influence measurements, particularly sensor offsets are critical for correct pattern recognition and subsequent motion data analysis.
In this project we aim to investigate the effect of misaligned sensors on the recognition of walking in free-living in patients after stroke. In particular we aim to analyse different supervised and unsupervised machine learning techniques to evaluate differences in recognition accuracy. We will simulate orientation offsets, implement and evaluate different algorithms including a rule-based unsupervised approach, and supervised approaches, e.g. one-class-classifier.
|Project type||Bachelor Thesis|
|Work distribution||20% Theory, 70% Programming and Data Analysis, 10% Experiment|
|Requirements||Signal processing, Python / MATLAB, Machine learning|
|Starting date||Summer term 2018|