Structure from Motion (SfM)

For more details - Project Report

Structure from Motion (SfM)

Implemented an end-to end pipeline for Structure from Motion to reconstruct a 3D scene from a set of images and simultaneously obtain the camera poses of the monocular camera with respect to the given scene. Steps involved Feature Matching and Outlier rejection using RANSAC, Estimating Fundamental using epipolar constraint and Essential Matrix, Estimate Camera Pose and Cheirality condition using Triangulation, PnP and Bundle Adjustment.

Input images

Set of monocular images used for 3D reconstruction and pose estimation.

”

Triangulation Check for Cheirality Condition

Initial 3D points are triangulated and validated using the cheirality condition.

”

Perspective-n-Points (PnP)

Camera poses are refined using PnP with 2D-3D correspondences.

”

Nonlinear Reprojection

Reprojection error minimized by refining poses and 3D points nonlinearly.

”

Bundle Adjustment

Global optimization of camera poses and 3D points to improve reconstruction accuracy.

”

Contributors: Mihir Deshmukh