Proposal of Artificial Mark to Measure 3D Pose by Monocular Vision

Abstract

To measure a 3D pose of the objects, we propose an artificial mark attached to objects. We propose three differently shaped marks, all of which have four point features. To measure the marks, the image processing system extracts the point features using edge extraction, erosion, and color extraction. Through measurement experiments, we compared error characteristic of these marks. The results showed that color extraction, which uses the simplest image processing method, was most robust when a mark was measured from various poses (mean of error: 10.4deg, mean of std. dev.: 7.06deg, worst std. dev.: 24.7deg) or was attached to the surface. Moreover, the mark enabled the shortest measurement time (0.478s/6marks). The usefulness of the mark with color extraction was verified through the success of an experiment in which a manipulator handled three marked objects.

Keywords

• artificial mark
• image processing
• measurement
• manipulator