Modelling Bipedal Locomotion Using Wavelets for Figure Animation

The thesis addresses the problem of how to develop a motion-modelling approach that can produce motion that is not only believable but also expressive and individual. In recent years motion capture techniques have been widely used for creating realistic animation of articulated figures. The realism obtained is difficult to achieve by other animation approaches. As motion capture supplies only sets of unstructured points representing joint trajectories, all information concerning the motion and the individual characteristics of the figure are hidden inside these clouds of data. This form of data may be sufficient if the expectation of the motion capture is simply to reproduce the motion. But if it is intended that the captured data should be easily reusable and editable and that the personal characteristic information should be extracted, then a proper functional representation of the captured data is essential. A new joint-modelling method for figure animation by using wavelets is explored in this thesis. The models adopted in this work depend upon captured data. To obtain a certain amount of motion data, a series of motion captured trials have been carried out using an optical system. The whole chain of using optical motion capture techniques from the physical set-up of the motion capture trial to ready-for-use captured data sets is presented in the thesis. Issues involved such as gathering joint position data of the markers as accurate as possible, removing noise without losing the secondary information hidden in the motion data and determining the segment lengths of the rigid skeleton from captured data are addressed. After an analysis of motion captured data, an underlying model is produced using wavelets to describe the angular variation of the joints. Instead of considering just individual frames, an entire motion process is treated as a whole. A wavelet model is built for each individual joint, and for each gait. Different motion sequences are expressed in terms of a common set of basis functions after that they are first normalised into a common reference frame. The work demonstrates that by using the multiresolution property of wavelets the joint motion curves can be decomposed hierarchically into successive levels of detail. In this way, localised detail can be separated from general trends. One can edit either the overall trend at some resolution to produce a different gait while retaining the style of the motion, or the details to produce a different character to the same motion. Editing the overall trend of the original motion is achieved by changing the values of only fewer “key points” at coarser resolution. Such an editing function supplies users a keyframing-like function model while keeping the burden of specifying the detail away from them. Editing the details can change the characteristics of the motion: this can be done by blending the “style” from other motion or the same individual in a different emotional state or even high frequency noise functions. The editing is not limited only to the above forms. In this work we have developed a new motion synthesis approach. By adjusting some high-level parameters - a set of weight factors, this synthesis operation blends the joint curves of two different motions to produce a family of different motions automatically. After editing or blending the existing motions, some constraints hidden in captured motion may be violated. It is important that in the final motion, the constraints are once again satisfied. The enforcement of both geometric constraints and time constraints are described in the thesis as well.

Name dc:type.qualificationname

PhD

Level dc:type.qualificationlevel

Doctoral

Grantor dc:publisher.institution

De Montfort University

Year dc:date.issued

2000

Author dc:creator

• Sun, Wei

Licence

• https://dora.dmu.ac.uk/bitstreams/22d2b3b2-7c96-42bf-a56a-f669a9104c45/download