In this paper we present a novel view point independent range image segmentation and recognition approach. We generate a library of 3D models off-line and represent each model with our tensor-based representation. Tensors represent local surface patches of the models and are indexed by a 4D hash table. During the online phase, a seed point is randomly selected from the range image and its neighbouring surface is represented with a tensor. This tensor is simultaneously matched with all the tensors of the library models using a voting scheme. The model which receives the most votes is hypothesized to be present in the scene. The model from the library is then transformed to the range image coordinates. If the model aligns accurately with a portion of the range image, that portion is recognized, segmented and removed. Another seed point is picked from the remaining range image and the matching process is repeated until the entire scene is segmented or no further library objects can be recognized in the scene. Our experiments show that this novel algorithm is efficient and it gives accurate results for cluttered and occluded range images.
3D Recognition and Segmentation of Objects in Cluttered Scenes
2005-01-01
751428 byte
Conference paper
Electronic Resource
English
Selective visual attention enables learning and recognition of multiple objects in cluttered scenes
British Library Online Contents | 2005
|A Scale Independent Selection Process for 3D Object Recognition in Cluttered Scenes
British Library Online Contents | 2013
|