Object Recognition with Imperfect Perception and Redundant Description

TL;DR

This paper presents a scene recognition system for robotics that matches images with human-provided, potentially redundant descriptions, effectively managing imprecision and uncertainty to improve object identification accuracy.

Contribution

It introduces a novel approach to handle imperfect perception and redundant descriptions in object recognition within robotic systems.

Findings

Effective management of imprecision and uncertainty in scene recognition

Enhanced matching likelihood through description redundancy assessment

Applicable to robotic object identification tasks

Abstract

This paper deals with a scene recognition system in a robotics contex. The general problem is to match images with <I>a priori</I> descriptions. A typical mission would consist in identifying an object in an installation with a vision system situated at the end of a manipulator and with a human operator provided description, formulated in a pseudo-natural language, and possibly redundant. The originality of this work comes from the nature of the description, from the special attention given to the management of imprecision and uncertainty in the interpretation process and from the way to assess the description redundancy so as to reinforce the overall matching likelihood.