Range information propagation transform

A novel method of model-based object recognition is presented in this paper. Its novelty stems from the fact that the gray level image captured by a camera is merged with sparse range information in an active manner. By using a projective transform, which is determined by the sparse range data, features (e.g. edge points) related to a single planar surface patch or figure in the scene can be assigned with their corresponding range values respectively. As a result, the shape of the very planar patch or figure can be recovered and various kinds of description in the Euclidean space can be calculated. Based on these descriptions values, the hypothesis about the identification of the object and its pose in space can be obtained with a high probability of success, and a high efficiency of hypothesis-verification process can be expected. Another advantage of this method is that the edge detection process can be navigated to the proper location hinted by the sparse range image. In consequence edge features can be extracted even in the regions with low contrast. In this paper the principle of range information propagation transform (RIPT) is explained, and some implementation issues, such as the algorithms using calibrated or uncalibrated gray level image for object recognition, are discussed. The preliminary experimental results are presented to indicate the effectiveness and efficiency of the proposed method.