Robust motion and 3D structure from space curves

Silhouettes of an object provide rich information about its geometry. They can be effectively used for shape reconstruction under known object movement. In this paper we propose a robust method to estimate the object rigid motion information from space curves using double stereo rigs. For any sequence, some convinced space curves are extracted from edge contours in each stereo rig images. The curve geometric distance, is minimized to track the space curves during the object movement. Double stereo setup is appeared to be robust against the statistical bias of motion translation parameters estimation. This setup doesnpsilat require the subpixel accuracy of edge pointpsilas location in stereo images. Besides, the curve matching scheme of the proposed method reduces the destructive effect of color maladjustment between cameras and illumination changes of light source in correspondence matching process. Experimental results demonstrate the quality of 3d shape reconstruction using the estimated motion information.