3D surface reconstruction from a single uncalibrated 2D image

Abstract

This paper described a simple computation to reconstruct 3D surface using a single uncalibrated 2D image from a digital camera as an image acquisition device that also focused on fast processing. An object is placed on a table with black background for the digital camera to shoot an image of the object. Image segmentation methods are applied in order to obtain the shape of the object from silhouette. The concept Radon transform is adopted to generate sinograms of the object and it is then inverse Radon transform is used to construct 2D cross-section of the object layer by layer. Canny edge detection helps to get the outline of each cross-section and coordinate points are extracted forming 3D point cloud from the image slices. 3D surface of the object is then reconstructed using Delaunay triangulation to connect each point with another. The results obtained from this project are satisfying regarding the processing time with recognizable shape and also strengthened with considerably low percentage error in the calculation for all six objects used in the experiment.