AR-Based 3D Scene Reconstruction from Multi-View Images
Project Overview
Implemented a complete AR-based 3D reconstruction pipeline using Structure-from-Motion principles. The system reconstructs a 3D point cloud from multi-view phone images using SIFT features, pose chaining, triangulation, and bundle adjustment.
Technologies Used
- C++ / OpenCV
- Ceres Solver for bundle adjustment
- SIFT Feature Detection
- Open3D for point cloud visualization
Key Features
- Custom pose chaining and triangulation logic
- Accurate camera calibration from phone EXIF data
- Experimental integration of MiDaS for dense depth fusion
- IEEE-format report and video presentation included
Research Contributions
- End-to-end SfM pipeline implemented from scratch
- Evaluation against COLMAP and MiDaS-based baselines
- Insights on 3D reconstruction accuracy from mobile sensors