Methods for Identifying Object Class, Type, and Orientation in the Presence of Uncertainty
Abstract
Techniques are presented for identifying unoccluded three-dimensional objects from arbitrary viewing angles in the framework of a model-based feature vector classification scheme. Fourier descriptors and moments are used for feature vector generation from contour imagery and silhouette and/or range imagery, respectively. A class of objects, airplanes, is defined with six distinct example types in our test data set. An additional data set of four objects from this class is also defined. A method for generating an exhaustive set of library views and worst case test views has been developed using a polyhedral approximation to a sphere. Based on matching to this library, object class membership, type, and orientation are determined. An approach called classification quality assessment (CQA) is applied to this recognition paradigm to both assess and deal with uncertainty. This is a two level process: the first rejects objects that are not members of a known class and therefore not contained in the model database, while the second identifies the likelihood of error for classification of known object type and/or orientation (within class errors). Both use simple measures that were generated solely from the system's priori knowledge.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1990
- Accession Number
- ADA225984
Entities
People
- Anthony P. Reeves
- Frank P. Kuhl
- Russell Taylor
Organizations
- United States Army Armament Research, Development and Engineering Center