University of Twente Student Theses


Activity classification through hidden Markov modeling

Tromper, T. (2016) Activity classification through hidden Markov modeling.

[img] PDF
Abstract:The goal of this study is to improve upon an existing activity classifier software model called FusionAAC, previously developed at Roessingh Research and Development. FusionAAC applies a technique called Hidden Markov Modeling to automatically and independently classify human activities. It mainly uses kinematic motion sensor data as input, but other input data types are also possible. The FusionAAC software is based upon the Hidden Markov Toolkit 3.4, implemented in a Labview environment. Until now, FusionAAC has been largely used as a 'black box' tool. This study looks to open up the black box by taking a closer look at the underlying algorithms. The deeper understanding is used to improve various aspects of the software, most notably the training process of the classifier. For the training process, the way the training data should be organized and different parameter initialization methods are discussed. Other areas of focus are preprocessing of the input data through application of principal components analysis, and the avoidance and removal of errors (most notably false positives) from the classification results. The improvements are tested on a data set containing various lifting activities.
Item Type:Essay (Master)
Roessingh Research and Development, Enschede, Netherlands
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:31 mathematics
Programme:Applied Mathematics MSc (60348)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page