Vögele, Anna Magdalena: Patterns in Motion - From the Detection of Primitives to Steering Animations. - Bonn, 2016. - Dissertation, Rheinische Friedrich-Wilhelms-Universität Bonn.
Online-Ausgabe in bonndoc: https://nbn-resolving.org/urn:nbn:de:hbz:5n-44845
@phdthesis{handle:20.500.11811/6891,
urn: https://nbn-resolving.org/urn:nbn:de:hbz:5n-44845,
author = {{Anna Magdalena Vögele}},
title = {Patterns in Motion - From the Detection of Primitives to Steering Animations},
school = {Rheinische Friedrich-Wilhelms-Universität Bonn},
year = 2016,
month = nov,

note = {In recent decades, the world of technology has developed rapidly. Illustrative of this trend is the growing number of affrdable methods for recording new and bigger data sets. The resulting masses of multivariate and high-dimensional data represent a new challenge for research and industry.
This thesis is dedicated to the development of novel methods for processing multivariate time series data, thus meeting this Data Science related challenge. This is done by introducing a range of different methods designed to deal with time series data. The variety of methods re ects the different requirements and the typical stage of data processing ranging from pre-processing to post- processing and data recycling.
Many of the techniques introduced work in a general setting. However, various types of motion recordings of human and animal subjects were chosen as representatives of multi-variate time series. The different data modalities include Motion Capture data, accelerations, gyroscopes, electromyography, depth data (Kinect) and animated 3D-meshes.
It is the goal of this thesis to provide a deeper understanding of working with multi-variate time series by taking the example of multi-variate motion data. However, in order to maintain an overview of the matter, the thesis follows a basic general pipeline. This pipeline was developed as a guideline for time series processing and is the first contribution of this work. Each part of the thesis represents one important stage of this pipeline which can be summarized under the topics segmentation, analysis and synthesis. Specific examples of different data modalities, processing requirements and methods to meet those are discussed in the chapters of the respective parts.
One important contribution of this thesis is a novel method for temporal segmentation of motion data. It is based on the idea of self-similarities within motion data and is capable of unsupervised segmentation of range of motion data into distinct activities and motion primitives.
The examples concerned with the analysis of multi-variate time series re ect the role of data analysis in different inter-disciplinary contexts and also the variety of requirements that comes with collaboration with other sciences. These requirements are directly connected to current challenges in data science.
Finally, the problem of synthesis of multi-variate time series is discussed using a graph-based example and examples related to rigging or steering of meshes. Synthesis is an important stage in data processing because it creates new data from existing ones in a controlled way. This makes exploiting existing data sets and and access of more condensed data possible, thus providing feasible alternatives to otherwise time-consuming manual processing.},

url = {https://hdl.handle.net/20.500.11811/6891}
}

Die folgenden Nutzungsbestimmungen sind mit dieser Ressource verbunden:

InCopyright