3-D Data Acquisition and Shape Recognition by Multiple Views

点击翻译按钮获取中文摘要

For recognition of 3-D shape and measurement of 3-D position it is important that a vision system can measure the 3-D data of dense points in the scene. One promising approach is to measure the distance on the basis of the triangulation principle from the disparity of two images. However, a problem of this binocular method is that it is not easy to find correspondence of feature points.In this research, we propose a new approach for this correspondence problem in which the correspondence problem can be solved geometrically by adding one more camera, that is, by trinocular vision. We have made the trinocular vision system and have made experiments on a variety of objects. From these experiments we have found that the geometrical constraints of the trinocular vision is very powerful for finding correspondence.However, when the density of edge points increases, chances of points which have multiple candidata points also increase. To solve this problem we also introduced other properties such as similarity of local properties and edge continuity. That is , when there are multiple candidates by geometrical constraints, we check similarity of local properties and edge continuity, and determine the true corresponding point.Using the above method, we have made experiments and have found that correct correspondence can be established even for complicated scene.