3D Modeling and Visualization By Morteza Daneshmand iCV Group, - PowerPoint PPT Presentation

3D Modeling and Visualization By Morteza Daneshmand iCV Group, Leader of the 3D Modeling and Computer Graphics Division Institute of Technology University of Tartu

Literature Review Contents Paper • Extracting 2D • landmark facial Automatic features facial feature • Computing the extraction and corresponding 3D coordinates 3D face • Estimation of the modeling using coordinates of the features hidden in two orthogonal the profile to align and locally views with deform the facial application to vertices 3D face recognition

Literature Review Contents Paper • Extracting 2D • Automatic landmark facial facial feature features • Computing the extraction and corresponding 3D 3D face coordinates modeling using • Estimation of the two orthogonal coordinates of the features hidden in views with the profile to application to align and locally deform the facial 3D face vertices recognition

Literature Review Contents Paper • Extracting 2D • Automatic landmark facial facial feature features • Computing the extraction and corresponding 3D 3D face coordinates modeling using • Estimation of the two orthogonal coordinates of the features hidden in views with the profile to application to align and locally deform the facial 3D face vertices recognition

Literature Review Paper • SUN3D: A Database of Big Spaces Reconstructe d using SfM and Object Labels

Literature Review Overview • Large-scale RGB- D video databases • Frames fully describing big scenes • Basic idea is useful for object reconstruction • Goal is to reconstruct a 3D point cloud

Literature Review Overview • Large-scale RGB- D video databases • Frames fully describing big scenes • Basic idea is useful for object reconstruction • Goal is to reconstruct a 3D point cloud

Literature Review Overview • Large-scale RGB- D video databases • Frames fully describing big scenes • Basic idea is useful for object reconstruction • Goal is to reconstruct a 3D point cloud

Literature Review Overview • Large-scale RGB- D video databases • Frames fully describing big scenes • Basic idea is useful for object reconstruction • Goal is to reconstruct a 3D point cloud

Literature Review Reconstructio n • Multiple Kinect frames of the scene are taken • The same idea could be used through taking frames of the same object from different orientations

Literature Review Reconstructio n • Multiple Kinect frames of the scene are taken • The same idea could be used through taking frames of the same object from different orientations

Literature Review Annotation • Camera poses • Object labels • Online object annotation • Structure from Motion (SfM) • Propagating the labels through the frames • This is not our concern right now!

Literature Review Annotation • Camera poses • Object labels • Online object annotation • Structure from Motion (SfM) • Propagating the labels through the frames • This is not our concern right now!

Literature Review Annotation • Camera poses • Object labels • Online object annotation • Structure from Motion (SfM) • Propagating the labels through the frames • This is not our concern right now!

Literature Review Annotation • Camera poses • Object labels • Online object annotation • Structure from Motion (SfM) • Propagating the labels through the frames • This is not our concern right now!

Literature Review Annotation • Camera poses • Object labels • Online object annotation • Structure from Motion (SfM) • Propagating the labels through the frames • This is not our concern right now!

Literature Review Annotation • Camera poses • Object labels • Online object annotation • Structure from Motion (SfM) • Propagating the labels through the frames • This is not our concern right now!

Literature Review Post- processing • Using object labels for error reduction • Generalized bundle adjustment • Object-to- object correspondence • Bounding boxes

Literature Review Post- processing • Using object labels for error reduction • Generalized bundle adjustment • Object-to- object correspondence • Bounding boxes

Literature Review Post- processing • Using object labels for error reduction • Generalized bundle adjustment • Object-to- object correspondence • Bounding boxes

Literature Review Post- processing • Using object labels for error reduction • Generalized bundle adjustment • Object-to- object correspondence • Bounding boxes

Methodology and Results The figure has been taken from http://www.cemyuksel.com/. Garment Model Reconstruction • Generating a 3D model of a garment from multiple shots • Preliminary works using ICP • 3D reconstruction of a cup • 3D reconstruction of a statue • 3D mesh based on the point cloud

Methodology and Results The figure is created by Lembit. Garment Model Reconstruction • Generating a 3D model of a garment from multiple shots • Preliminary works using ICP • 3D reconstruction of a cup • 3D reconstruction of a statue • 3D mesh based on the point cloud

Methodology and Results Garment Model Reconstruction • Generating a 3D model of a garment from multiple shots • Preliminary works using ICP • 3D reconstruction of a cup • 3D reconstruction of a statue • 3D mesh based on the point cloud

Methodology and Results Garment Model Reconstruction • Generating a 3D model of a garment from multiple shots • Preliminary works using ICP • 3D reconstruction of a cup • 3D reconstruction of a statue • 3D mesh based on the point cloud

Methodology and Results Garment Model Reconstruction • Generating a 3D model of a garment from multiple shots • Preliminary works using ICP • 3D reconstruction of a cup • 3D reconstruction of a statue • 3D mesh based on the point cloud

Methodology and Results Existing Challenges • Too few points • Minimum and maximum distances from the camera • Noisy data and wrong scales • Missing data • Loop closure • Local minimums • Fallacious transformation • Adjusting thresholds

Methodology and Results Existing Challenges • Too few points • Minimum and maximum distances from the camera • Noisy data and wrong scales • Missing data • Loop closure • Local minimums • Fallacious transformation • Adjusting thresholds

Methodology and Results Existing Challenges • Too few points • Minimum and maximum distances from the camera • Noisy data and wrong scales • Missing data • Loop closure • Local minimums • Fallacious transformation • Adjusting thresholds

Methodology and Results Existing Challenges • Too few points • Minimum and maximum distances from the camera • Noisy data and wrong scales • Missing data • Loop closure • Local minimums • Fallacious transformation • Adjusting thresholds

Methodology and Results Existing Challenges • Too few points • Minimum and maximum distances from the camera • Noisy data and wrong scales • Missing data • Loop closure • Local minimums • Fallacious transformation • Adjusting thresholds

Methodology and Results Existing Challenges • Too few points • Minimum and maximum distances from the camera • Noisy data and wrong scales • Missing data • Loop closure • Local minimums • Fallacious transformation • Adjusting thresholds

Methodology and Results Existing Challenges • Too few points • Minimum and maximum distances from the camera • Noisy data and wrong scales • Missing data • Loop closure • Local minimums • Fallacious transformation • Adjusting thresholds

Methodology and Results Existing Challenges • Too few points • Minimum and maximum distances from the camera • Noisy data and wrong scales • Missing data • Loop closure • Local minimums • Fallacious transformation • Adjusting thresholds

Thank You