Evolution of Point Cloud Extraction Techniques from SFM to SLAM and Their Applications in 3D Reconstruction

Abstract

This study aims to address the issue of computational time efficiency in traditional Structure from Motion (SFM) for sparse point cloud and pose estimation. Traditional SFM techniques provide valuable three-dimensional scene information in multi-view geometry reconstruction, but their computation speed is relatively slow, limiting their widespread application in real-time scenarios. To enhance computational speed, this research adopts Simultaneous Localization and Mapping (SLAM) technology as a replacement for traditional SFM methods. SLAM not only generates sparse point cloud and pose information more rapidly but also accomplishes the goal of real-time Structure from Motion in unknown environments. By fusing sensor data and motion models, SLAM technology demonstrates significant improvements in time efficiency in multi-view geometry reconstruction. The results of this study indicate that using SLAM as an alternative to SFM can significantly reduce computation time without compromising reconstruction quality, making it more suitable for applications requiring real-time performance, such as autonomous driving and robotic navigation. This innovative approach provides a new time-efficient solution for the field of 3D reconstruction, expanding its application prospects across various domains.