Reconstruction of the 3d Object Model: a Review

The three-dimensional (3D) reconstruction model of a real object is useful in many applications ranging from medical imaging, product design, parts inspection, reverse engineering to rapid prototyping. In the medical field, imaging techniques such as computed tomography (CT), magnetic resonance imaging (MRI) and single positron emission tomography (SPECT) are applied to create 3D images from emanation measurements for disease diagnoses and organ study. On the other hand, reconstruction is widely utilized to redesign manufacturing parts in order to save production cost and time. A typical reconstruction application consists of three major steps, which are data acquisition, registration and integration as well as surface fitting. Based on the nature of data captured, the 3D reconstruction model can be categorized into two groups: methods working on (i) two-dimensional (2D) images and (ii) sets of 3D points. This paper reviews different methods of 3D object model reconstruction and techniques subjected to each method. 1.0 Introduction Reconstruction is a process of rebuilding by new construction the exact form and detail of a vanished structure, object or a part thereof as it appeared in as a specific period of time. In early time, it has been widely applied in medical imaging. CT, MRI, SPECT, 3D Ultrasound (US) and positron emission tomography (PET) provide a wide range of 3D medical imaging modalities. In industry various models for the automatic reconstruction of objects are proposed [Ghassan et al., 2007; Ng et al., 2009]. The need for the 3D reconstruction model is motivated by several factors. The manual creation of the 3D model is time consuming and therefore expensive (Lim & Shamsuddin, 2004). Besides, the increasing availability of 3D scanning devices has responded to this need. Generally, the reconstruction procedure encompasses four (4) steps [Lim & Shamsuddin, 2004] (Figure 1). Data acquisition is the process of sampling physical objects to be reconstructed. Sampling devices can be coordinated measuring machines (CMM) that generate sets of points, cameras that produce intensity images or a stereoscopic image, acoustic devices such as sonar that generate height maps and medical scanning devices such as CT and MRI scanner that generate volumetric information. The most common sampling device today is the laser scanner that generates range data. Registration refers to the process where the range data are aligned to transform all the measurements into a common Euclidean space [Montani et al., 2004]. In the integration stage, all the data are merged into a single surface model. The aim of surface fitting is to construct a concise and accurate approximation of the physical surface. The display stage denotes the visualization of the final model representation by using a rendering technique. The objective of this