This thesis addresses the problem of automatic measurement of human motion in a multiple-camera studio environment, without use of markers or special clothing. Potential applications include: computer animation from video sequences, human-computer interaction, biomechanical and ergonomic analysis. Existing pose estimation schemes are reviewed, and some implementations of these are evaluated. The existing schemes have drawbacks in terms of ability to track complex motion, deal with ordinary clothing, track multiple people, and recover from errors. Such problems are found to stem from deficiencies in some of the search schemes, and the simple models of appearance which are often used. To address these issues, a new hierarchical stochastic sampling search scheme is introduced, together with a more detailed model of human appearance. The new method is found to be effective for a range of complex motions, a variety of normal clothing, and simultaneous pose estimation of multiple people. A new fast wand-based method for calibration of multiple camera studios is also given, and its performance is found to compare favourably with an existing chart-based approach.