Improving the performance of video based reconstruction and validating it within a Telepresence context

Duckworth, TW

Abstract

This thesis investigates visual, spatial and temporal qualities of video based reconstruction with respect to telepresence. State of the art was improved and validated through a new parallelisation of an established algorithm; a tool that allows visio-spatial impact of algorithm and camera arrangement to be visualised; and a set of experiments to derive requirements and investigate outcomes. The motivation is to support the exchange of appearance and attention between moving humans through video based reconstruction. A previous research project showed moving humans could faithfully convey attention in virtual environments and appearance through video-conferencing, suggesting that it may be possible to combine the two. Video based 3D reconstruction of humans appeared to be able to achieve both, but it was uncertain whether this could be achieved at sufficient quality. Research began by justifying the approach and setting the requirements. A literature survey and initial experiments indicated that the visual hull provided a form suitable for modeling humans. However, evidence of visual and temporal qualities necessary to support gaze was not found. A state of the art visual hull reconstruction algorithm was parallelised to run on a modern multi-core processor, enabling human reconstruction on a single computer, thus providing stable visual and temporal qualities. A parallelisation scheme theoretically better suited for execution on a single multi-core processor than distributed over a network is proposed. Importantly, the way in which the problem is parallelised has been optimised to reflect the various stages of the process rather than the need to minimise data communication across a network. A utility application has been developed providing a framework for rapidly testing algorithms, validating requirements, and as a platform for conducting experiments. This underpinned a collaborative experiment that showed for the first time that eye gaze could be conveyed to accuracies sufficient for human social interaction. To facilitate the analysis, the utility allowed the impact of camera placement on spatial and visual quality to be investigated.

Citation

Duckworth, T. Improving the performance of video based reconstruction and validating it within a Telepresence context. (Thesis). University of Salford

• Thesis Type: Thesis
• Deposit Date: Jul 5, 2013
• Publicly Available Date: Apr 5, 2016
• Additional Information: Funders : Engineering and Physical Sciences Research Council (EPSRC)