Image analysis – Pattern recognition – Template matching
Reexamination Certificate
1999-03-04
2001-07-10
Johns, Andrew W. (Department: 2621)
Image analysis
Pattern recognition
Template matching
Reexamination Certificate
active
06259815
ABSTRACT:
FIELD OF THE INVENTION
This invention relates generally to computer vision, and more particularly to recognizing scanned physical objects using deformable three-dimensional templates.
BACKGROUND OF THE INVENTION
A number of techniques are known for recognizing three-dimensional objects from two-dimensional images of the objects. Brooks, in “Model-based 3D interpretation of 2D images,” IEEE Trans. on Pattern Analysis and Machine Intelligence, 5(2):140-150, 1983, describes the use of parameterized objects for object recognition. There the objects are generalized cylinders.
Solina et al., in “Recovery of parametric models from range images,” IEEE Trans. on Pattern Analysis and Machine Intelligence, 12(2):131-146, 1990, describe how to recover parametric objects from range data by minimizing surface distances, for example, using superquadrics with global deformations. There is no object recognition here.
Dickenson et al. survey a number of different shape recovery techniques in “From volumes to views: An approach to 3D object recognition,” Computer Vision, Graphics and Image Processing: Image Understanding, 55(2):130-154, 1992.
Most prior art template matching methods do not use full volumetric templates. The templates are either two-dimensional for image matching, or in the case of range data, the templates are surface based. In addition, the prior art templates are relatively unconstrained to maximize the likelihood of matching. Unconstrained templates require the manipulation of a large number of parameters.
Therefore, it is desired to provide a vision-based recognition system that can identify physical objects having multiple constituent parts. Furthermore, it is desired to parse the identified objects so that they can be animated in virtual environments.
SUMMARY OF THE INVENTION
Provided is a computer vision system for recognizing and parsing common toy-like objects, such as people, animals, trees, houses, cars, boats, birds, insects, etc. A toy object can be made of modeling clay, or other suitable materials. Using a single calibrated camera, a sequence of images of the model object is acquired while the object is rotated on a turn-table. A 3D volumetric scan of the object is recovered from the sequence of 2D images.
The volumetric scan is compared against sets of three-dimensional deformable templates. There is one set of three-dimensional deformable templates for each category of object. This process allows one to both classify the object and to identify its constituent parts. Using this information, it becomes possible to recognize and parse, for example, an ambulatory object, such as a biped or a quadruped, and to place the object in an appropriate virtual environment for animation.
More particularly, a method recognizes three-dimensional physical objects using three-dimensional deformable templates. A particular object is scanned with a camera to generate volumetric data representing the object. The volumetric data is compared to each of a plurality of three-dimensional deformable templates stored in a database to obtain a score for each comparison. The templates are deformed while the score of an objective function is optimized.
The three-dimensional deformable template that receives the highest score determines the category of the object, and the template is used to parse the object into its constituent parts, and the parts of the object can than be rendered and animated in a virtual world environment.
REFERENCES:
patent: 5768413 (1998-06-01), Levin et al.
patent: 5926568 (1999-07-01), Chaney et al.
patent: 6009212 (1999-12-01), Miller et al.
Rodney A. Brooks; “Model-Based Three-Dimensional Interpretations of Two-Dimensional Images”;IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-5. No. 2, Mar. 1983, pp. 140-150.
Sven J. Dickinson & Dimitri Metaxas; “Integrating Qualitative and Quantitative Shape Recovery”; International Journal of Computer Vision,, 5(2) :311-33, 1983.
John C. Russ; “Euclidean Distance Map”;The Image Processing Handbook, 2ndEdition; pp. 463-469.
Franc Solina & Ruzena Bajcsy; “Recovery of Parametric Models from Range Images: The Case for Superquadrics with Global Deformations”;IEEE Transactions on Pattern Analysis and Machine Intelligence, 12(2) :131-146, 1990.
Agarwala Aseem
Anderson David B.
Beardsley Paul
Marks Joseph W.
Azarian Seyed
Brinkman Dirk
Johns Andrew W.
Mitsubishi Electric Research Laboratories Inc.
LandOfFree
System and method for recognizing scanned objects with... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with System and method for recognizing scanned objects with..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and System and method for recognizing scanned objects with... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2485935