The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

Research Paper #752

Title:Non-Polyhedral Landmark Recognition Using 3D Depth Images and Partially Correct Models
Authors:Fillatreau,P; Fisher,RB
Date:Jun 1995
Presented:To appear in the Proceedings of the 3rd International Symposium on Intelligent Robotic Systems (SIRS'95), Italy, 10-14 July 1995
Abstract:In this paper, we address the problem of modeling, recognising and localising non-polyhedral shapes or objects in partially known scenes, using 3D depth images. The applications this work aims at are mobile robotics for intervention or survey in environments like factories or nuclear plants. This paper focusses on the problems of the recognition of non-polyhedral landmarks or objects and misplaced feature detection. In our approach, planes and some simplified classes of quadrics play an important part in the different processes. Surface classification takes into account the uncertainties on the measures and relies on probabilistic tests. The surface features belonging to those simplified classes provide a pose estimate which can be refined, if needed, using iconic methods. After registration, we compare the scene description and then a priori model for model validation and change detection.

[Search These Pages] [DAI Home Page] [Comment]