Fast and robust 3D textured model reconstruction on mobile platforms

Abstract

We describe a fast and robust 3D modeling framework in mobile environments with a RGB-D camera. The proposed framework estimates the camera pose using a small number of keypoint patches from the color images and builds a 3D volume using the two-layer voxel hashing structure from the depth images with the estimated pose. It is also robust to large motions by using the keypoint patches of the rendered image from the predicted camera pose and the integrated volume. As demonstrated in the experiments, our framework runs on the mobile platform and generates high-quality 3D models in real-time.