Vision based 3D mapping-From Traditional to NeRF based approaches

3D reconstruction or 3D mapping of an environment is one of the most crucial stages of Simultaneous Localisation and Mapping (SLAM). Numerous work have been done to optimize the tracking and mapping process of SLAM systems over the years in both classical computer vision and deep learning fields. Although there have been many surveys that extensively study SLAM-based work, most of them do not discuss 3D mapping and its developments in much detail. In this paper, we discuss the history of SLAM from a general perspective as well as focus on 3D reconstruction/mapping. To our knowledge, our paper is the first to dedicatedly explore Neural Radiance Field (NeRF) research that is used for SLAM, pose estimation and 3D reconstruction. Thus we track the history of mapping techniques in classical feature-based, direct-based, deep learning-based and most importantly NeRF based literature. Finally, we make a comparative study of all the existing methods and discuss the challenges faced by these concluding the survey.