Object Modelling of Multiple Views Using Dual Quaternion in Reverse Engineering

Abstract

The common requirement in reverse engineering, as well as in computer vision is to calculate the 3D rigid transformations that exist among different data sets acquired in multiple views. Of the many existing methods, the ICP (iterative closest point) is the most popular one. However, the traditional ICP method still has certain drawbacks. In this paper an improved method is presented which is called IDCP (iterative dual closest point). The improvements are demonstrated from three aspects. One is the data organisation form. We design a novel structure which is called adaptive roulette. It can adjust the area of its cell according to the number of points and put data points into different cells. This form can make it efficient to search the nearest neighbouring point. The second is the core method IDCP. When we construct the points pair we not only use the minimum Euclidean distance rule, but also refer to the comparability of normals of the points concerned. This will increase the accuracy and decrease the iteration times significantly. The third is integration of the two view pairs. We give a way to integrate the two triangle sets by avoiding the conflict in topology of modelling.