Stereo-vision for autonomous industrial inspection robots

Abstract

Camera-based stereo-vision provides cost-efficient vision capabilities for robotic systems. The objective of this paper is to examine the performance of stereo-vision as means to enable a robotic inspection cell for haptic quality testing with the ability to detect relevant information related to the inspection task. This information comprises the location and 3D representation of a complex object under inspection as well as the location and type of quality features which are subject to the inspection task. Among the challenges is the low-distinctiveness of features in neighboring area, inconsistent lighting, similar colors as well as low intra-class variances impeding the retrieval of quality characteristics. The paper presents the general outline of the vision chain as well as performance analysis of various algorithms for relevant steps in the machine vision chain thus indicating the capabilities and drawbacks of a camera-based stereo-vision for flexible use in complex machine vision tasks.