Vision systems are going to be ever-present in many different applications and studying the factors which influence their accuracy is of high interest in the computer vision community. In most of the applications of camera-based 3D scanners, the medium between the camera and the object is air, whereas in some specific applications such as sole scanning, the medium is a flat see-through glass. A piece of glass can influence the accuracy of the scanner due to the refraction of light it causes; therefore, in this work, the effect of glass on the camera calibration step, which is intended to recognize the parameters of a vision system, is studied. For this purpose, a setup is developed and the calibration object is captured through pieces of glass with different thicknesses and distances from the camera and the calibration algorithm is run for different series of images. Reprojection Error, as a common criterion for evaluating the accuracy of the camera calibration process, is used for analyzing the effect of glass. The results demonstrate that the effect of the studied thicknesses of glass is very small keeping the average reprojection error less than two pixels and can be interpreted as ignorable in most applications.
Paper
Full text
A Study on the Refractive Effect of Glass in Vision Systems
Semantic Scholar · Physics · 2019
Abstract
Vision systems are going to be ever-present in many different applications and studying the factors which influence their accuracy is of high interest in the computer vision community. In most of the applications of camera-based 3D scanners, the medium between the camera and the object is air, whereas in some specific applications such as sole scanning, the medium is a flat see-through glass. A piece of glass can influence the accuracy of the scanner due to the refraction of light it causes; therefore, in this work, the effect of glass on the camera calibration step, which is intended to recognize the parameters of a vision system, is studied. For this purpose, a setup is developed and the calibration object is captured through pieces of glass with different thicknesses and distances from the camera and the calibration algorithm is run for different series of images. Reprojection Error, as a common criterion for evaluating the accuracy of the camera calibration process, is used for analyzing the effect of glass. The results demonstrate that the effect of the studied thicknesses of glass is very small keeping the average reprojection error less than two pixels and can be interpreted as ignorable in most applications.