Robust 3-D Object Reconstruction Based on Camera Clustering With Geodesic Distance

This letter presents a method that reconstructs a three-dimensional (3-D) object using camera clustering and key camera selection to resolve the scalability problem. To perform camera clustering, camera similarity is defined using the geodesic distance and overlap constraint between cameras. Key cameras are then selected to reconstruct an object considering overlapping cameras and high curvature regions. As a result, it is achievable to relax increases in the execution time and the accumulated errors due to the scalability problems with a large number of cameras.