Toward motion-capture-based digital human modelling

Abstract: Digital Human Models (DHMs) are getting more applicable in the simulation of human postures/actions inside virtual production systems. Despite the existence of various modeling techniques, simulated motions are still looking unrealistic. One way to increase the natural looking of simulated motions is to use real human data as a source for motion simulation algorithms. Motion capture is a common tool to record real human motions and has been widely used in animation and game applications. It is observed that although many research studies embrace the profitability of using real human data for motion simulation in DHM tools, the existing motion capture data can be hardly reused in a systematic way. Number of reasons such as variations in skeleton configurations and motion formats, and non-efficient annotating of motions are identified as the reasons for this problem. In this thesis first a data schema for motion capture data management is presented. The purpose of such schema is to store and manipulate motion data in a way that mentioned problems are not arising. A synthesizer platform is also presented which is able to store motions taken from real human in the database and synthesize new motions using already existing motions. The platform is able to search the database, analyze the data, and feed the data to DHM tools. The platform functionality was tested within two areas: a) by composing new motions by combining the arm motions and walking motions from different subjects and b) by using analysis tools to compare generated motions in a commercial software against captured motions from real human.