The author presents a general method for determining the three-dimensional position and orientation of an object relative to a camera based on a two-dimensional image of known feature points located on the object. In contrast to the conventional approach, the method does not make use of the collinearity condition. Instead, the algebraic structure of the problem is fully exploited in arriving at a solution which is independent of the configuration of the feature points. Although the method is applicable to any number or distribution of feature points, no more than five points are needed from a numerical standpoint; three or four points will typically suffice. A necessary condition for the existence of the solution is obtained for the first time, and a rigorous proof of uniqueness is presented for the case of four coplanar points. Simulation results have shown that, in the case of four feature points, nonplanar configurations generally outperform coplanar point sets in both accuracy and robustness in the presence of image noise.
展开▼
