Monocular Structure from Motion

Abstract

Structure from motion (SFM) is the construction of a 3-D view of a scene. The technique uses a series of 2-D images. The project aims at the 3-D reconstruction of the environment/scene by taking images with a single camera inmotion thus getting the useful information and dimensions of the environment. In this project, we implemented a general pipeline for structure from motion by working on an opensource code coded in python programming language. The pipeline implements image processing techniques to obtain key pointsfrom 2-D images, matching those key points with all otherimages ofsame scene taken from different angles and then creating its 3-D view in the point cloud form. In this project we tried and tested different algorithms of computer vision to obtain the best key points and their matches and reported our results. We also came up with the best possible algorithm to minimize the reprojection error that occurs during 3-D reconstruction.