It is a very important problem to extract accurate depth information in 3-D object recognition because the features used in the recognition should be estimated from the depth information. But it is not possible to extract accurate depth information in real situation because of various reasons including variation in brightness, in light source direction, and so on. Therefore, the statistical properties of features due to estimation error of depth information should be considered.
This thesis analyzes the statistical properties of features using those of depth information estimated by means of stereo method. And then an adaptive object recognition algorithm which considers statistical properties of features according to condition that object is placed on 3-D space is proposed. Several experimental results show the usefulness of the proposed algorithm.