On the Recovery of Quadric Surface Shapes via Simultaneous Boundary and Surface Fitting

Abstract

本文提供的技巧，可以用來從對應的光度及景深影像中，偵測出二度物體的子部位形狀。主要的假設包括:(1)物體最突顯的特徵乃是發生在物體的輪廓部位以及物體表面上的邊線，(2)物體可能有互相遮掩的情形，故重建物體的形狀只能藉重於物體較小的部位。本文所提供的方法，乃是經由將線條及表面的資料，同時調適，使得以匹配二次曲面形式的一程技巧。所須要使用的圓球、圓柱及圓錐二次曲面結合邊線的方程式在文中均有推導。藉蒙地卡羅方式的測驗，我們得以證實:(1)線條及表面資料同時調通的結果，以調適的品質來看，會使於個別調適的方式;(2)利用對應光度及景深影像資料來偵測物體的形狀，可以降低物體表面錯誤分類的發生。

A method is proposed for detecting the shape of small parts ofobjects from fused intensity and range data. My primary assumptionsare: (1) the most telling object features occur on the object rim andat object creases, and (2) occlusion existing in complex scenes requiresthe reconstruction of sensed objects in terms of these small parts. Afitting method that simultaneously fits specific quadric forms to bothsurface and contour data at object limbs or creases is presented. Fittingequations for spherical, cylindrical and conical limbs and for polyhedralcreases are derived. Results from Monte Carlo experiments support thehypotheses that (1) fitting both surface and contour is superior, interms of the quaality of the fit, to either surface fitting or contourfitting alone; (2) the shape of the surface is less likely to bemisclassified when we use the proposed fused data fitting method.