System and method for recognition in 2D images using 3D class models
First Claim
1. A method for recognizing class instances in a 2D image, comprising the steps of (a) constructing a database of 3D class models;
- and (b) matching class instances appearing in the 2D image to the 3D class models, by identifying one or more image features in the 2D image and comparing the identified image features with class parts belonging to the 3D class models.
1 Assignment
0 Petitions
Accused Products
Abstract
A system and method for recognizing instances of classes in a 2D image using 3D class models and for recognizing instances of objects in a 2D image using 3D class models. The invention provides a system and method for constructing a database of 3D class models comprising a collection of class parts, where each class part includes part appearance and part geometry. The invention also provides a system and method for matching portions of a 2D image to a 3D class model. The method comprises identifying image features in the 2D image; computing an aligning transformation between the class model and the image; and comparing, under the aligning transformation, class parts of the class model with the image features. The comparison uses both the part appearance and the part geometry.
153 Citations
33 Claims
-
1. A method for recognizing class instances in a 2D image, comprising the steps of
(a) constructing a database of 3D class models; - and
(b) matching class instances appearing in the 2D image to the 3D class models, by identifying one or more image features in the 2D image and comparing the identified image features with class parts belonging to the 3D class models. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
- and
-
17. A database of 3D class models, where:
-
(a) each 3D class model comprises a plurality of class parts;
(b) each class part comprises a part geometry and a part appearance; and
(c) each part appearance comprises a first part appearance descriptor that is computed from 2D information. - View Dependent Claims (18, 19, 20)
-
-
21. The system of 19, where the second part appearance descriptor represents the canonical appearance of the corresponding elements.
-
22. A method for recognizing class instances in a 2D image, using a database of 3D class models, comprising the steps of
(a) identifying one or more image features in the 2D image, each feature having a feature location and a feature appearance; -
(b) constructing one or more correspondence hypotheses, each correspondence hypothesis comprising a class model and a set of feature-part matches;
(c) using each correspondence hypothesis to compute an aligning transformation of the class model to the 2D image;
(d) evaluating each correspondence hypothesis under the aligning transformation by computing a class score; and
(e) choosing correspondence hypotheses whose class scores meet selected acceptance criteria.
-
-
23. A system for recognizing class instances in a 2D image, comprising
(a) a database of 3D class models; (b) a computer configured to match class instances appearing in the 2D image to the 3D class models, by identifying image features in the 2D image and comparing the image features with class parts belonging to the 3D class models. - View Dependent Claims (24)
-
25. A computer-readable medium storing information for performing the steps of:
-
(a) constructing a database of 3D class models; and
(b) matching class instances appearing in the 2D image to the 3D class models, by identifying one or more image features in the 2D image and comparing the identified image features with class parts belonging to the 3D class models.
-
-
26. A method for recognizing objects in a 2D image, comprising the steps of:
-
(a) constructing a database of 3D object class models where each 3D object class model comprises a plurality of object class parts, each object class part comprising a part geometry and a part appearance, where each part appearance comprises a first part appearance descriptor that is computed from 2D information and a second part appearance descriptor that is pose invariant; and
(b) matching object instances appearing in the 2D image to the 3D object class models, by identifying one or more image features in the 2D image and comparing the identified image features with object class parts belonging to the 3D object class models. - View Dependent Claims (27, 28, 29)
-
-
30. A method for recognizing deformable objects in a 2D image, comprising the steps of:
-
(a) constructing a database of 3D object class models where each 3D object class model comprises a plurality of object class parts, each object class part comprising a part geometry and a part appearance, where each part appearance comprises a first part appearance descriptor that is computed from 2D information and a second part appearance descriptor that is pose invariant, where the part geometry models variability of the object due to deformations; and
(b) matching object instances appearing in the 2D image to the 3D object class models, by identifying image features in the 2D image and comparing the image features with object class parts belonging to the 3D object class models.
-
-
31. A method for recognizing objects in a 2D image, where the objects have multiple appearance states, comprising the steps of:
-
(a) constructing a database of 3D object class models where each 3D object class model comprises a plurality of object class parts, each object class part comprising a part geometry and a part appearance, where each part appearance comprises a first part appearance descriptor that is computed from 2D information and a second part appearance descriptor that is pose invariant, where one or more of the part appearance descriptors model the multiple appearance states of the object; and
(b) matching object instances appearing in the 2D image to the 3D object class models, by identifying image features in the 2D image and comparing the image features with object class parts belonging to the 3D object class models.
-
-
32. A system for recognizing objects in a 2D image, comprising:
-
(a) a database of 3D object class models where each 3D object class model comprises a plurality of object class parts, each object class part comprising a part geometry and a part appearance, where each part appearance comprises first appearance descriptor which is computed from 2D information and a second part appearance descriptor that is pose invariant. (b) a computer configured to match object instances appearing in the 2D image to the 3D object class models, by identifying one or more image features in the 2D image and comparing the identified image features with object class parts belonging to the 3D object class models.
-
-
33. A computer-readable medium storing information for performing the steps of:
-
(a) constructing a database of 3D object class models where each 3D object class model comprises a plurality of object class parts, each object class part comprising a part geometry and a part appearance, where each part appearance comprises a first part appearance descriptor that is computed from 2D information and a second part appearance descriptor that is pose invariant; and
(b) matching object instances appearing in the 2D image to the 3D object class models, by identifying one or more image features in the 2D image and comparing the identified image features with object class parts belonging to the 3D object class models.
-
Specification