抄録
This paper proposes a method to recognize poses of figures which have arbitrary locations, sizes and rotations for measuring poses of two-dimensional industrial components. A template figure is represented by line segments that consist of the edge directions and curvature ratios of a pair of pixels. In the matching process, the values in the addresses that mean the candidates for pose parameters are accumulated by using the invariable geometric parameters for pose recognition including in line segments. The location of objective figures are extracted at the peak addresses in the voting space, and the sizes and the rotation angles are obtained by relative poses included in the voting log. The experimental results, show that the poses of objective components were recognized successfully even if the components were overlapped.
