Comparison of Three Dimensional Measurement Accuracy Using Stereo Vision

Abstract

The stereo vision system has been widely used in recent years. The imaging process of a camera consists in projection from a three dimensional world to a two dimensional image in which one dimension of information has been lost. A vision system that consists of two cameras can use the implicit geometry constraint between the images to recover the three dimensional information. Thus, the stereo vision system can be used as a measuring tool. A few applications have been reported and some of them are included by the literature review. In this thesis, two parameters of the stereovision system have been used to test the measuring results: the distance between the cameras and the distance between the cameras and the object. Experimental results show that when the object is close to the cameras (about 912mm), three dimensional measurement has relatively small error. The error increases when the distance becomes larger. Experiments also verify that the larger the distance between the cameras (within the tested range), the better the measuring results, except for the measurements along the depth direction. A quantitative analysis is used to analyze the measuring sensitivity to the movement of the corner point. It shows that measurement along the axis is affected the least. The comparison also verifies that when the distance between the cameras becomes larger, the system becomes less sensitive to the small movement of the corner point.