Research Seminar - August 08, 2003

Automated 3-D Model-based Object Recognition

Abstract:

Object modeling and recognition find their application in computer graphics, virtual reality, industrial design, robotics and automated surgery etc. The first step in 3-D modeling is data acquisition. Data is acquired using various techniques like stereo, shape from shading and triangulation etc, which gives us data in the form of a depth image. A complex 3-D object cannot be modeled with a single view and requires multiple views, which are later on registered into a common coordinate system. Present techniques used for registration are not fully automatic and require tracking of the acquisition system or human intervention for finding correspondences between various views. Registration is followed by integration and reconstruction to complete the 3-D model. The 3-D model is then stored in a model library using a representation scheme. During the recognition phase a view of an object is acquired and converted into a similar representation scheme for matching with the model library. Present representation schemes are non-unique and are limited to a small domain.

I propose a fully automated 3-D model-based object recognition scheme capable of automatic 3-D modeling of objects and recognition in occluded and cluttered environments.