Abstract
Virtual humans are the computer representation of real humans in virtual worlds. Human representation in virtual worlds is vital as people are central elements of the real and virtual worlds. Animating virtual humans remains a challenging task due to the large number of degrees of freedom in the human skeleton, the ability of human observers to detect unnatural movements, and the complexity of realistic appearance and behaviour. The difficulty of the problem has resulted in research into many subproblems of human animation synthesis. In this thesis, we focus on the motion control problem. The objective is to synthesise realistic human motion with improved control over the resulting animation. Motion capture of real actors provides a source of realistic animation sequences. Many existing motion editing techniques require an expert or trained animator to deal with complex models or to define the desired motion by manipulating low-level variables. A few high-level editing techniques have been recently introduced. However, their common drawbacks include the need of relatively large database of animation sequences and the high computation complexity. This thesis introduces a high-level parametric approach for synthesis of realistic human animation based on synchronised blending of existing motion clips. The proposed approach allows intuitive control of the desired motions while generating realistic animation based on a small motion captured database of basic motion clips. A framework has been developed and utilised to realise interesting and challenging tasks such as parametric synthesis of animation sequences along arbitrary paths and over uneven terrains. The extension for control of multiple parameters simulteneously is also presented. Other contributions have been made to allow the synthesis of long animation sequences from short animation clips which reduces the input data requirements and costs. The presented approach, based on motion blending, is computationally efficient with linear computation complexity with length of the desired animation sequence. It also requires a relatively small database of basic motion clips. Visual inspection is the ultimate evaluation of animation. However, a number of quantitative measures are introduced and adopted in this thesis to provide consistent and systematic evaluation of the resulting animation. Results demonstrate realistic synthesis of novel motion sequences using intuitive parameters such as speed, direction, and slope.