3D Reconstruction Of the Human Torso From Four Orthogonal Model Views Captured Using a Stationary RGB-D Camera
MetadataShow full item record
Digital 3D models of the human body can be used to measure anthropometric features to diagnose diseases. Circumference measurements of body parts can only be obtained from 360-degree spin 3D models. These type of models are created using a combination of scanning hardware and stitching software. Current stitching techniques use an Iterative Closest Point algorithm on manually or automatically generated registration landmarks, provided that there is a rotation angle difference of 60 degrees or less between adjacent frames for sufficiently matching features. This paper proposes a method to stitch four orthogonal (90-degree) views captured using one Kinect camera by iteratively translating and rotating adjacent frames to minimize the Root Mean Square Error between the contours of the torso in 2D planes. Future work will involve mending holes in the stitched models and solving the shear transformation to improve results on non-rigid bodies.