On calibrating computer controlled cameras for perceiving 3-D scenes

摘要

A problem that arises is getting computers to perceive 3-D scenes is relating information from several different viewpoints. In particular, if the computer moves its sensor, it has to be able to predict changes in images of objects it has seen without having to completely re-recognize them. A solution to this problem has been implemented at Stanford using a calibrated camera model which expresses the relation between object space and image space as a function of the computer's control variables. The modelling problem is relatively well understood. Calibration techniques, however, are not. This article deals with these.