3-D TRANSLATIONAL MOTION AND STRUCTURE FROM BINOCULAR IMAGE FLOWS

In this paper, image flow fields from parallel stereo cameras are anal yzed to determine the relative 3-D translational motion of the camera platform with respect to objects in view and to establish stereo corre spondence of features in the left and right images. A two-step procedu re is suggested. In the first step, translational motion parameters ar e determined from linear equations whose coefficients consist of the s ums of measured quantities in the two images. Separate equations are d eveloped for cases when measurements of either the full optical flow o r the normal flow are available. This computation does not require fea ture-to-feature correspondence. In addition, no assumption is made abo ut the surfaces being viewed. In the second step of the calculation, w ith the knowledge of the estimated translational motion parameters, th e binocular flow information is used to find features in one image tha t correspond to given features in the other image. Experimental result s with synthetic and laboratory images indicate that the method gives accurate results even in the presence of noise.