3D invariant estimation of axisymmetric objects using fourier descriptors

摘要

Three-dimensional (3D) invariants are important for 3D object recognition and analysis. Lin and Jungthirapanich [Pattern Recognition 23, 833–842 (1990)] have suggested the use of contour-related 3D elliptic Fourier invariants for object recognition. Since the contour of an object depends on the view angle, recognition of an object can be unreliable. In this research, the 3D Fourier descriptor (FD) based volume-type (V-type) and area-type (A-type) invariants are proposed and a method to estimate these invariants for axisymmetric objects is developed. These invariants have the property of being invariant with translation, rotation and scaling. With the proposed invariants, object recognition as well as object property analysis for mass, density and dynamics can be exercised. Experiments are conducted for a cylinder, a concentrically cylindrical object and a wine bottle. The results show that the proposed V- and A-type invariants remain unchanged with respect to location, orientation, scaling and view direction. Further, as the number of harmonics reaches 29, percentage of error for estimating the volume and the area reduces to 2%, with the computation time less than 0.1 s on a PC486/33