Real Time 3D Reconstruction from Monocular Video

Real Time Monocular 3D Reconstruction

This book summarizes the knowledge acquired along the development activities performed along the writing and development of my PhD Thesis research. It is written using the Markdown language but extended with citation support from a BibTeX database using Pandoc, and compiled with GitBook to simplify editing and publishing. The goal of this book is to serve as a kind of diary of my research activities. Here I will take notes of my experiments and lessons learned along my research.

3D reconstruction from a single video stream (often called monocular camera) recovers the depth of scene objects using information from consecutive frames. My PhD aims to improve the accuracy and performance of the overall process introducing a feedback between the reconstruction and the recognition stages.

At a first step, the research is focused only on interior scenes with closed loops using triangulation between different camera positions at a graph. Energy minimization algorithms will be applied when possible in optimization and regularization procedures.

Results could be applied to medicine, architecture, forestry, robotics or audiovisual production, among others.