Monocular Image-based 3D Structure Generation
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.
Implemented robust feature detection and matching algorithms for reliable correspondence across multiple views
Computed fundamental matrices and essential matrices for geometric relationship estimation between camera views
Applied Perspective-n-Point algorithms and bundle adjustment optimization for accurate camera pose estimation
Designed and integrated complete reconstruction pipeline from feature extraction to 3D model generation
80% contribution as lead researcher, driving technical decisions and implementation
Successfully built end-to-end reconstruction system from monocular inputs
Achieved reliable 3D structure generation across various building types
OpenCV, SIFT, SURF, Bundle Adjustment
C++, Python, MATLAB
Linear Algebra, Optimization, Geometry
PCL, Eigen, Ceres Solver