Interactive learning of a multiple attribute hash table for fast three-dimensional object recognition

Lynne L Grewe, Purdue University

Abstract

Multiple-attribute hashing is now considered to be a powerful approach for the recognition and localization of 3D objects on the basis of their invariant properties. In the systems developed to date, the structure of the hash table is fixed and must be created by the system developer--an onerous task especially when the number of attributes is large, as it must in systems that use both geometric and non-geometric attributes. Another deficiency of previous systems is that uncertainty is treated as a fixed value and not modeled. We have created a system, named MULTI-HASH, which uses the tools of decision trees and uncertainty modeling for the automatic construction of hash tables. The decision-tree framework in MULTI-HASH is based on a hybrid method that uses both qualitative attributes, such as the shape of a surface, and quantitative attributes such as color, dihedral angles, etc. The human trainer shows objects to the vision system and, in an interactive mode, tells the system the model identities of the various segmented regions, etc. Subsequently, the decision-tree based framework learns the structure of the table. Another contribution of this research is the successful integration of geometric and non-geometric attributes in a coherent framework. A special sensor, a color structured light scanner, was constructed for the registration of photometric information with the 3D range data of a scene. The implemented system has been used to successfully and efficiently guide a robot in recognizing and removing objects from piles, in the presence of occlusion and against cluttered backgrounds. We show the results of these experiments and other comparative results.

Degree

Ph.D.

Advisors

Kak, Purdue University.

Subject Area

Electrical engineering|Engineering|Computer science

COinS