Inferring Motion of Cylindrical Object from Shape Information

Abstract

This paper explores methods for Inferring motion of a cylindrical object from an image sequence. A local shading analysis method segments each Image Into spherical, cylindrical, and planar 'surfaces. The cylindrical surface Is characterized with a direction of generating lines determined from spatial derivatives In the image. An extended reflectance map method Is applicable to estimate gradients of base In consecutive frames If the lighting conditions are known. Another approach to Inferring of motion Is to use shape Information which needs an additional constraint. If two planes of the object are hot parallel, then the finding of correspondence between frames Is easy. For the case of parallel planes, we introduce an additional assumption that the planes are orthogonal to the generating lines. This strong constraint Is not sufficient for determining the surface normals from single view. The orthogonality also gives another constraint between a change In area of the base plane and that In length of generatling fine from the first to the second frames. Thus Iwe can estimate orientations of the bcise plane from two Views, Finally, motion parameters are estimated by correlating two shapes of the base boundary mapped from the two views.