Camera Calibration from Human Motion (2004-2005)
With two or more 2D views of a 3D scene, we can reconstruct the scene and extract information of interest but only if we know where the cameras are with respect to each other. Determining the relative positions of the cameras (plus their individual parameters such as zoom) is known as calibration and typically makes use of a static object of known dimensions.
In this work, however, we looked at using things we know about the human body to calibrate the camera. In particular, we know that the body is symmetric and that individual bones don't grow and shrink from one second to the next. Using these properties, we calibrated two cameras more efficiently and accurately than previous methods under similar conditions.