Stereo Method Using Local Direction Histogram for Reference Points Teaching

In off-line teaching of a robot, motion data is generated on model space in a computer. The data is then to be converted to the coordinate system in real working space. This paper presents a simple stereo method for this coordinate transformation. The stereo method runs as follows. Each stereo image is divided into small rectangle sub-areas for fast feature extraction. In each area, a local direction histogram (LDH) is calculated from intensity gradient. By the shape of LDH, sub-areas which include vertices of the object are selected as feature areas. Then 3-D location of vertices are calculated by stereo-corresponding of these feature areas. The transformation matrix between model space and real working space is obtained by selecting corresponding vertices as reference points from stereo data and from the object model. The method was implemented on a personal computer. A result of simulation experiment of reference points teaching is shown.