Vision-Based Measurement of Part Deformation and Misalignment for Deformable Cylindrical Peg-in-Hole Tasks

Abstract For successful assembly of deformable parts, information about their deformation and possible misalignments between the holes and their respective mating parts is essential. Such information can be mainly acquired from visual sensors. In this paper, part deformation and misalignment in cylindrical peg-in-hole tasks are measured by using a visual sensing system. First, the configuration and the specifications of the system, such as resolution, are described. Next, a series of experiments to measure the position of an arbitrary point are performed and its measurement accuracy is investigated. Then, an algorithm to estimate the centre-line and deformation of a cylindrical peg and an algorithm to divide and recognize a peg and a hole separately in an image are presented. On the basis of these algorithms, a series of experiments to measure part shape as part deformation are performed. Finally, an algorithm to select two views from the four on the image plane and an algorithm to estimate the centre of an occluded hole are presented. On the basis of these algorithms, a series of experiments to measure misalignment are performed. Experimental results show that the errors in measuring part deformation are approximately less than five or seven times the standard resolution of the system, and the errors in measuring misalignment are less than three or four times the standard resolution. Thereby, the system and the proposed algorithms are effective in measuring part deformation and misalignment and will dramatically increase the success rate in deformable assembly operations.