Part-based Construction of digitized 3D objects

Abstract

Nowadays, a few 3D acquisition devices are available at low-cost. While 3D capture is a commonplace, decompose the object into its components is not an easy task. Segmentation can help address this problem by suppling data which may be used to identify object components. However, it might not give complete and accurate information about components. In a context where a digital repository with every component that can belong to physical objects is available, retrieval algorithms can be used to construct a composed 3D model. We propose a four phase solution to construct 3D digitized objects. We use Microsoft Kinect® to acquire 3D physical objects. A segmentation algorithm based on color information decomposes the object into a set of sub-parts. The component repository is queried using a shape-based retrieval algorithm, in order to identify which sub-part corresponds to each virtual component. Then, a 3D model of the physical object is constructed by assembling the retrieved components. The work presented in this paper has a wide application domain, ranging from entertainment to health or mechanical industry. To validate our proposal, we implemented a toy-problem and evaluated its precision and efficiency. We used LEGO® blocks, which can provide challenges similar to real-world applications. The results were encouraging and we believe that our approach may even work better with greater object components, geometrically less similar to each other.