| This thesis is the result of my graduate research carried out at the junction point of the following three major areas: (1) volumetric shape and movement analysis of articulated objects; (2) high-performance multi-perspective visual computing; (3) advanced methods of applied mathematics for large scale problems.; The work was motivated by a set of vision problems that arise in multi-perspective environments. The primary focus was on shape and movement recovery of complex articulated objects such as human bodies.; To begin, I present a real-time volumetric shape recovery system based on visual cone intersection that runs on a multi-sensor PC cluster.; Next, I study complex shapes with density based models, develop two frameworks for 3D articulated shape hierarchical approximation, and show how one can use them for human body shape and movement analysis.; Finally, I introduce a flexible multi-perspective vision studio that flexible plug-in based system of multi-view vision procedures, and provides a 3D scene viewer.; In addition, I present an efficient cross-platform implementation of Generalized Fast Multi-pole Method, whose C++ code relied solely on portable standard libraries and showed excellent performance on large scale problems of acoustics, applied physics and (scientific) data visualization. |
