3D object recognition using coded light projection for robot assembly applications

Abstract

This paper describes the algorithms needed to allow automatic recognition of components in three dimensions (3-D) using a novel active range sensor based on liquid crystal technology, which is programmed to project coded lighting sequences onto a pallet of workpieces within a flexible manufacturing cell. A range map with up to 1200/spl times/1200 range points is constructed using a sequence of binary image captures from a single robot head camera. Calibration of the vision sensor head is achieved using a grid of black squares on a white mat placed on the pellet prior to on-line operation and accuracy of 0.33% in height measurement are achieved over a working area of 100/spl times/100 cm. The range data is processed on a digital signal processing image analyser and construction solid geometry (CSG) surfaces are matched to the data sets to extract surface models of the components. The surface model is stored in an object database and used to recognise the position of workpieces in the robot coordinate frame.