A common setting in Activity Recognition is that sensors, such as triaxial accelerometers, are worn on the body or embedded into objects of daily use. The recorded multi-variate sensor streams undergo analysis in order to infer the activities that were performed by the user. Simple yet effective methods, such as k-NN classification using statistical features, often suffice to obtain impressive recognition accuracies. Therefore information about what subjects are doing is readily available, rendering activity segmentation a straight-forward followup task. However, so far relatively little work was invested into a further, detailed analysis of these segmented activities, although extracting their characteristics, i.e. how well these activities were performed, would be beneficial to a variety of applications spanning many domains.
My main research goal is to develop novel methods for the assessment of this motor skill, particularly for applications in medicine. Here many degenerative conditions such as Parkinson’s Disease and Dementia have a significant impact on motor abilities, where motor assessment is crucial for early intervention and treatment.