Probabilistic Approach to Modeling of 3D Objects using Silhouettes

Abstract

Recently, 3D model construction from 2D images using an uncalibrated camera has attracted significant attention in the research community. Most of the algorithms for 3D model construction suffer from problems such as inefficiency, irregular construction, and necessity of camera calibration. In this paper, a novel algorithm is presented that uses the silhouette images obtained from the object to construct the 3D model. To carry out the 3D modeling, multiple views of the object are taken from different angles. Then using a silhouette based technique, new silhouettes are constructed and feature points are derived from them. These feature points are then used to construct the triangular meshes, which in turn construct the whole surface of the 3D model. The noise in the silhouette images is dealt with a probabilistic framework. In addition, a faster technique is presented to reduce the time and space complexity of this algorithm making it feasible for most commercial applications. The algorithm has been successfully tested on several objects. The experimental results and comparison with a voxelization technique over several sequences shows the superiority and the effectiveness of our technique.