A vision system with 3D capabilities

摘要

The first results of a vision system which adds to a 2D image the depth information of certain points are presented. The goal of the system is to provide the control unit of a robot with an approximate 3D description of the industrial part to be manipulated or of the working area. Depth information about the selected points of the scene is obtained by stereo vision. The determination of disparity is solved by changing the point of view of the TV camera employed by means of two oscillating mirrors which allow continuous tracking of the selected points. A specialized image processor has been designed to deal with the tasks of selection of outstanding points of objects (vertices) and continuous tracking of these points between consecutive images when displacing the mirrors in order to obtain a short time of calculation as well as a low cost system. That way at the end of each sweep of the mirrors we get the information needed to solve triangulation from pairs of homologe points. The time spent to get the three dimensional information of the scene is about one second, which corresponds to the time need to get enough intermediate images to guarantee the correct tracking of vertices.