Depth Map Fusion with Camera Position Refinement Computer Vision - PowerPoint PPT Presentation

Introduction Method Experiments Conclusion Depth Map Fusion with Camera Position Refinement Computer Vision Winter Workshop 2009 cek and Radim ˇ Radim Tyleˇ S´ ara tylecr1@cmp.felk.cvut.cz Center for Machine Perception Department of Cybernetics, Faculty of Electrical Engineering, Czech Technical University Prague, Czech Republic cek and Radim ˇ Radim Tyleˇ S´ ara, CMP Prague Depth Map Fusion, CVWW 2009

Introduction Method Experiments Conclusion Contents Introduction 1 Reconstruction Pipeline Method 2 Idea Algorithm design Depth task Visibility task Experiments 3 Results Evaluation Future work Conclusion 4 cek and Radim ˇ Radim Tyleˇ S´ ara, CMP Prague Depth Map Fusion, CVWW 2009

Introduction Method Reconstruction Pipeline Experiments Conclusion 3D Reconstruction Pipeline Input ⇒ Corresponding Disparity Surface ⇒ ⇒ Point cloud ⇒ images regions maps mesh Pair-wise vs. multi-view stereo Calibration inaccuracy Inconsistency between disparity maps cek and Radim ˇ Radim Tyleˇ S´ ara, CMP Prague Depth Map Fusion, CVWW 2009

Introduction Method Reconstruction Pipeline Experiments Conclusion Input 3D point cloud Noise, outliers in the data Redundancy cek and Radim ˇ Radim Tyleˇ S´ ara, CMP Prague Depth Map Fusion, CVWW 2009

Introduction Idea Method Algorithm design Experiments Depth task Conclusion Visibility task Surface representation Reconstructed surface (scan) Depth maps ⇒ Back- Surface composition projection Visibility maps Linear complexity cek and Radim ˇ Radim Tyleˇ S´ ara, CMP Prague Depth Map Fusion, CVWW 2009

Introduction Idea Method Algorithm design Experiments Depth task Conclusion Visibility task Idea of Depth Map Fusion Representation with a set of reference cameras supporting camera scene 1 supporting camera 4 1−2 2−4 supporting camera 3 fused depth map 2−3 2 pairwise disparity map reference camera cek and Radim ˇ Radim Tyleˇ S´ ara, CMP Prague Depth Map Fusion, CVWW 2009