Skip to main content
🏗️ Computer Vision

3D Building Exterior Reconstruction

Monocular Image-based 3D Structure Generation

KETIAug 2020 - Dec 2020

🎯 Project Overview

Developed a comprehensive 3D reconstruction module capable of generating detailed 3D building structures from monocular images. This project involved implementing state-of-the-art computer vision algorithms including keypoint matching, epipolar geometry computation, and bundle adjustment optimization for accurate 3D scene reconstruction.

🔬 Technical Implementation

SIFT/SURF Keypoint Matching

Implemented robust feature detection and matching algorithms for reliable correspondence across multiple views

Epipolar Geometry

Computed fundamental matrices and essential matrices for geometric relationship estimation between camera views

PnP & Bundle Adjustment

Applied Perspective-n-Point algorithms and bundle adjustment optimization for accurate camera pose estimation

3D Pipeline Integration

Designed and integrated complete reconstruction pipeline from feature extraction to 3D model generation

🚀 Key Achievements

🎯

Lead Researcher Role

80% contribution as lead researcher, driving technical decisions and implementation

🏗️

Complete 3D Pipeline

Successfully built end-to-end reconstruction system from monocular inputs

Robust Performance

Achieved reliable 3D structure generation across various building types

💡 Technical Highlights

  • Advanced keypoint matching with RANSAC outlier rejection for robust correspondences
  • Multi-view geometry optimization using bundle adjustment for global consistency
  • Efficient camera calibration and pose estimation pipeline
  • Dense point cloud generation and mesh reconstruction algorithms
  • Real-time visualization and quality assessment tools

🛠️ Technologies Used

Computer Vision

OpenCV, SIFT, SURF, Bundle Adjustment

Programming

C++, Python, MATLAB

Mathematics

Linear Algebra, Optimization, Geometry

Tools

PCL, Eigen, Ceres Solver

🔗 Resources