The k1k2 space in range image analysis

Range finder images may be segmented by using a decision space H(k1, k2) where k1 represents the maximum local surface curvature and k2 the minimum surface curvature. Since clusters within H(k1, k2) represent various types of surfaces, the pixels in the range image Z(x, y) are classified accordingly. Homogeneous regions within Z(x, y) which correspond to classes in H(k1, k2) are thus automatically recognized according to their surface curvature characteristics. The advantages of using the maximum-minimum surface curvature (k1, k2) decision space for the first classification step in range image processing and understanding are demonstrated.<<ETX>>