Over the past several years, interest in computer simulated cloth algorithms has grown steadily. Recent advances in simulation algorithms allow complex and realistic looking cloth to be simulated for use in the motion picture industry. Unfortunately, it remains quite difficult to adjust the parameters of these simulations to match a given real world cloth material.
It may be possible to acquire motion data for moving cloth, and subsequently recover the cloth simulation parameters that best fit this motion data. A recent paper [11] investigated this possibility, but used only 3D geometry data for the recovery. Better simulation parameter recovery would be possible if motion data contained information about stretching, shearing and bending of the cloth, and if occlusions due to folds could be better understood.
In this thesis, we present a new method for measuring the 3D geometry and surface parameterisation of moving cloth, captured from a stereo video sequence. Stretching, shearing and bending could be calculated from the data we measure, and it may also be possible to recover cloth simulation parameters.
In Chapter 2, the previous work is discussed including papers in cloth simulation, cloth parameter recovery and cloth motion capture. Chapter 3 contains an analysis of the influence of simulator parameters on a popular cloth simulation algorithm. The subsequent chapters deal with various stages of the new cloth motion capture algorithm, as shown in Figure 1.1. Chapter 4 describes the disparity map construction, Chapter 5 describes the recovery of a surface parameterisation of the cloth, and Chapter 6 demonstrates the results obtained by the cloth motion capture algorithm. The thesis closes with some conclusions and future work in Chapter 7.
Some of this material has been published previously in a different form. Chapters 5 and 6 were originally part of a paper entitled Cloth Motion Capture [88] written with Wolfgang Heidrich.