3D Reconstruction From Traditional Methods to Deep Learning

Vision is one of the important pathways for human perception of external information, with over 80% of perception being acquired through vision. How to enable computers to possess efficient and flexible visual systems similar to humans has always been a hot topic in the field of computer science. One of the main goals of computer vision research is to reconstruct the geometric structure of 3D objects visible on the visible surfaces from 2D photos. Recently, this technology has become mature enough and its applications range from autonomous driving, virtual reality, cultural heritage preservation and restoration, among others, with significant research value. In this paper, we summarize the key technical issues in 3D reconstruction from existing technologies, first by summarizing traditional methods of 3D reconstruction, then analyzing commonly used deep learning methods for 3D reconstruction and their application scenarios in different fields. Finally, we conclude and provide an outlook on future development directions.