COSMOS: A Framework for Representation and Recognition of 3D Free-Form Objects

Abstract: Free-form object recognition, also referred to as sculpted object recognition, is an emerging theme of importance in the field of computer vision. This dissertation presents a new approach to automated representation and recognition of 3D free-form rigid objects using dense surface data. We describe a computer vision system that recognizes arbitrarily curved 3D rigid objects from a single view when (a) the viewpoint can be arbitrary, (b) the objects may vary in shape and complexity, and (c) no restrictive assumptions are made about the types of surfaces on the objects. We assume that a range image of a scene is available, and contains a view of a rigid 3D object without occlusion. Availability of CAD models of 3D objects, although not a necessity, is considered an advantage and exploited in our system to easily generate multiple views of the object for its model construction. Our surface representation scheme, COSMOS, describes an object concisely in terms of maximal surface patches of constant shape index. These maximal patches that represent the object are mapped onto the unit sphere via their orientations, and aggregated via shape spectral functions. Surface properties such as area, curvedness, and connectivity which are required to capture local and global information about the object are also built into the representation. The scheme yields not only a meaningful and rich description useful for the recoverability of several classes of objects, but also provides a set of powerful matching primitives for recognition. We present a recognition strategy which consists of a multi-level matching mechanism employing shape spectral analysis and features derived from the COSMOS representations of objects for fast and efficient object identification and pose estimation. Shape spectra based object view representations are employed for efficient view grouping and model organization with large databases. Given a range image of an uncluttered view (allowing self-occlusion) of an object, the shape spectrum based model selection scheme short-lists a few promising candidate views from a database of object views. The COSMOS-based view verification scheme then establishes the correct object identity of the input by comparing the COSMOS representations of the views in detail using a ``patch-group graph'' matching technique. Estimation of the pose of the recognized object is formulated as registration of the sensed data with the range image of the best matched view of the object. We present a minimum variance estimator to robustly register two range images of a complex object and compute their relative view transformation accurately. All theoretical aspects of this work have been experimentally validated via a prototype system, which has been tested on a database of over 6,000 object views generated from CAD models and surface triangulations and 100 range images of several complex objects acquired using a 3D laser range scanner.

Key words: 3D Free-form objects, arbitrary shapes object , representation, COSMOS, shape spectrum, view grouping, recognition, view matching, patch-graph matching, pose estimation, registration.

Figures...

Related Publications:

" Recognition of Free-Form Objects" Chitra Dorai and Anil K. Jain , Proceedings of the 13th International Conference on Pattern Recognition, Vienna, Austria, Vol. I, pp. 697-701, August 1996.

" Registration and Integration of Multiple Object Views for 3D Model Construction" Chitra Dorai, Gang Wang, Anil K. Jain and Carolyn Mercer, IEEE Trans. Patt. Anal. Mach. Intel., vol. 20, no. 1, pp. 83-89, January 1998. A short version of this paper, "From Images to Models: Automatic 3D Object Model Construction from Multiple Views," appeared in the Proceedings of the 13th International Conference on Pattern Recognition, Vienna, Austria, Vol. I, pp. 770-774, August 1996.

" Shape Spectrum Based View Grouping and Matching of 3D Free-Form Objects" Chitra Dorai and Anil K. Jain, IEEE Trans. Patt. Anal. Mach. Intel., vol. 19, no. 10, pp. 1139-1146, October 1997. A short version of this paper, "View Organization and Matching of Free-Form Objects," appeared in the Proceedings of the IEEE International Symposium on Computer Vision, Coral Gables, Florida, pp. 25-30, November 1995.

" Shape Spectra Based View Grouping for Free-Form Objects" Chitra Dorai and Anil K. Jain, Proceedings of the International Conference on Image Processing, Washington, D.C., Vol. III, pp. 340-343, October 1995.

"COSMOS---A Representation Scheme for 3D Free-Form Objects" Chitra Dorai and Anil K. Jain, IEEE Trans. Patt. Anal. Mach. Intel., vol. 19, no. 10, pp. 1115-1130, October 1997. A short version of this paper, "COSMOS---A Representation Scheme for Free-Form Surfaces," appeared in the Proceedings of the fifth International Conference on Computer Vision, Boston, Massachusetts, pp. 1024-1029, June 1995.

"Optimal Registration of Object Views Using Range Data " Chitra Dorai, John Weng and Anil K. Jain, IEEE Trans. Patt. Anal. Mach. Intel., vol. 19, no. 10, pp. 1131-1138, October 1997. A short version of this paper, "Optimal Registration of Multiple Range Views," appeared in the Proceedings of the 12th International Conference on Pattern Recognition, Jerusalem, Israel, pp. 569-571, October 1994.

For other publications click here