The stereo vision system endows a robot with intelligence and flexibility since some three-dimensional properties can be extracted from it. In order to apply the 3-D vision to the robotic operations, the vision system should satisfy requirements such as the real-time processing, portability and low cost implementation. The conventional stereo algorithms for 3-D vision are, however, very complex and time consuming. The purpose of this research is to develop a simple and computationally efficient stereo algorithm which is suitable for applications to robotic operations.
The concept of the superimposed difference image is introduced in the proposed algorithm. It is shown that, under certain conditions, the features exist in the form of pairs in the superimposed difference image. The disparities and the depth map are determined by searching the paired features within the searching window. The complex stereo matching problem is avoided since only the superimposed difference image is processed to extract the features.
The depth map is used to obtain three-dimensional properties of objects. The object is approximated by a hexahedron and the vertices of the hexahedron are obtained by detecting the extreme points of the top view and the side view which are projections of the depth map onto the plane. The occupied space of object is determined for obstacle detection of robot, and the three-dimensional properties such as the 3-D centroid, size and orientation are determined for the robotic manipulation such as the pick-up of an object.
The experiments are performed in a laboratory setting and the results for depth map and three-dimensional properties of objects are obtained satisfactorily. The analysis for the noise effect and a comparative study with a typical stereo algorithm are also performed.