Vision can be used to position a robot relative to a known object or a known environment in 3D. If the object has enough feature points, one view is sufficient for determining the relative position between the object and the camera, otherwise, multiple views are required. We discuss the mathematics of viewpoint determination, using a combination of calibration matrix decomposition and space resection. The combined method has low noise sensitivity and does not require knowledge of camera parameters. If the object does not have enough features, multiple views are required to determine its position and orientation; an example of this is be given. The formulation of homogeneous transform equations to drive the manipulator to the goal position is also be given.
Date of this Version