We developed a number of human motion tracking methods. All our methods take their input from multiple-camera systems.

The first method that we developed takes as input silhouettes gathered with several synchronized and calibrated cameras. The method uses kinematic and geometric models of the human body. The image silhouettes can be interpreted as the extremal contours of a curved surface ... (details).

The second method takes as input 3D data such as surface patches with their position and orientation. The method uses an articulated implicit surface to represent the human body. The data-to-surface registration process is carried out probabilistically using a variant of the expectation-maximization algorithm ... (details).