Visual Analysis of Multidimensional Data for Biomechanics and HCI

Multidimensional analysis is performed in many scientific fields. Its main tasks involve the identification of correlations between data dimensions, the investigation of data clusters, and the identification of outliers. Visualization techniques often help in getting a better understanding. In this thesis, we present our work on improving visual multidimensional analysis by exploiting the semantics of the data and enhancing the perception of existing visualizations. Firstly, we exploit the semantics of the data by creating new visualizations which present visual encodings specifically tailored to the analyzed dimensions. We consider the resulting visual analysis to be more intuitive for the user as it provides a more easily understandable idea of the data. In this thesis we concentrate on the visual analysis of multidimensional biomechanical data for Human-Computer Interaction (HCI). To this end, we present new visualizations tackling the specific features of different aspects of biomechanical data such as movement ergonomics, leading to a more intuitive analysis. Moreover, by integrating drawings or sketches of the physical setup of a case study as new visualizations, we allow for a fast and effective case-specific analysis. The creation of additional visualizations for communicating trends of clusters of movements enables a cluster-specific analysis which improves our understanding of postures and muscular co-activation. Moreover, we create a new visualization which addresses the specificity of the multidimensional data related to permutation-based optimization problems. Each permutation of a given set of n elements represents a point defined in an n-dimensional space. Our method approximates the topology of the problem-related optimization landscape inferring the minima basins and their properties and visualizing them organized in a quasi-landscape. We show the variability of the solutions in a basin using heat maps generated from permutation matrices. Furthermore, we continue improving our visual multidimensional analysis by enhancing the perceptual encoding of existing well-known multidimensional visualizations. We focus on Parallel Coordinates Plots (PCP) and its derivative Continuous Parallel Coordinates (CPC). The main perceptual issues of PCP are visual clutter and overplotting which hamper the recognition of patterns in large data sets. In this thesis, we present an edge-bundling method for PCP which uses density-based clustering for each dimension. This reduces clutter and provides a faster overview of clusters and trends. Moreover, it allows for a fast rendering of the clustered lines using polygons. Furthermore, we present the first bundling approach for Continuous Parallel Coordinates where classic edge-bundling fails due to the absence of lines. Our method performs a deformation of the visualization space of CPC leading to similar results as those obtained through classic edge-bundling. Our work involved 10 HCI case studies and helped to establish a new research methodology in this field. This led to publications in internationally peer-reviewed journals and conference proceedings.