Towards improving performance and user-friendliness of optical motion capture systems
Measuring the motion of humans, animals or objects is important in various current and future applications. Today, many alternative motion capture systems exist to obtain such measurements. The technological approach to detect motion is often used to classify the systems. For instance, optical systems were among the first ones applied successfully, and are to date the preferred choice when high precision is required. While the current technology is mature, recent research indicates that better understanding of high-precision optical systems can improve their performance and user-friendliness. The purpose of this thesis is to explore different research directions related to such improvements.
We start off with an introduction to motion capture from a historical and technological perspective: from the pioneering works using analog photography in the late 19th century to a comparison of different technologies employed in present-day solutions. We next present a possible break-down of the typical motion capture problem that modern systems have to tackle into six stages: Preparation, Measurement, Reconstruction, Tracking, Identification and Post-Processing. We focus specifically on how marker-based optical systems usually approach these stages.
Drawing upon knowledge from industry experts and academic literature, we compile a set of stage-specific and system-level topics in marker-based optical motion capture where better understanding is anticipated to elucidate ways of improving performance and user-friendliness. From these, we select a subset of topics in the Preparation stage to focus on as part of this thesis. We employ an ensemble of methods from photogrammetry, human-computer interaction (HCI) and visualization in our exploratory research. In particular, we conduct an initial investigation into geometry-aware projectors that could, amongst others, be used to support camera placement; we compare two alternative lens-distortion models and assess the robustness of a global optimization algorithm used in camera calibration; further, we propose a dynamic simulation method to visualize residual errors from camera model calibration; for a wand calibration we suggest a wand-mounted feedback device to guide the user; finally, we present the main challenges of introducing test automation to an existing signal-processing software systems. While conclusive validation was not performed for any of these topics, the initial results contributed by this thesis do indicate that there is a need for better understanding in each of them.
The thesis concludes with a detailed description of two real-world applications the author was involved in to varying degrees. We summarize the lessons learned from these involvements.
Visual Arena, Lindholmen, Chalmers University of Technology
Opponent: Prof. Jürgen Gall, Institute of Computer Science III, University of Bonn, Germany