3D scene reconstruction using a single monocular image

• Main Authors: Ablay, Francis Miguel P., Agustin, Mark Christian P., Cengca, Marc Lorenz M., Tiam-Lee, Thomas James Z.
• Format: text
• Language: English
• Published: Animo Repository 2012
• Subjects: Pattern Recognition; Three-dimensional imaging; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/10651
• Institution: De La Salle University

This is research on 3D scene reconstruction from a single monocular 2D image. It extends the study done by Saxena, Sun, and Ng (2008) called Make3D. Make3D constructs a 3D model from a single 2D image using monocular cues extracted from it. In this study, additional features not used by Make3D have been identified and used to produce a better 3D reconstruction. These features are the sky blob, saturation, and object blob. These features are extracted from the image and used in the improvement of the 3D model. Qualitative survey was used to evaluate the outputs of this study on unstructured images. Results of the survey reveal that this study has produced better results in some cases when compared against Make3D.