Using Machine Vision to Command a 6-axis Robot Arm to Act on a Randomly Placed Zinc Die Cast Product

Abstract

The paper presents a method, documenting the machine vision techniques required, to automate a manual process at a local New Zealand manufacturing company. The system was required to monitor a conveyor belt for zinc die cast outputs. A robot arm was informed how to pick up the object. Die cast outputs appear face up or down and can be rotated 360°. Four experiments were conducted to determine how accurately the proposed system could detect if the cast was lying face up or down, determine the robot picking location, and determine the angle of direction while performing an error checking function. Circle Hough transform and principal component analysis, along with other object analysis techniques, were used on 2D datasets. Point cloud data was analysed to determine if the cast was face up or down when taken from a time-of-flight camera.