3D Reconstruction of Long Bone using Kinect

This paper proposes the new framework of the 3D reconstruction of the long bone structure using Kinect. It consists of three main steps including capturing, segmentation and alignment of the point clouds from multiple views. In the capturing process, the setting is empirically designed to achieve the complete information of the long bone. Then, the segmentation process is to separate the background from the foreground object i.e., the long bone. In the last step, the segmented point clouds of the long bone from multiple views are aligned to form up the corresponding 3D model. It is the double-layer alignment process. The first layer is based on the geometric information taken in the capturing process. The second layer is to refine the output from the first layer, using the alignment of color markers seen under consecutive views. Based on the experiment, the proposed method is shown to have a promising performance, by comparing key measurements of the reconstructed model with the manual ground truth.