In this thesis, we propose a 3D reconstruction method based on passive sensing techniques by integrating multiple visual cues. Intensity images from multiple viewpoints are taken as input data. Multiple visual cues such as volumetric intersection, occluding contours and stereo contribute together in the process of 3D reconstruction. Deformable superquadrics are chosen as geometric primitives for object representation. Physics-based approach is adopted to simulate the 3D reconstruction process, where the visual modules cooperate with each other. More importantly, the proposed method is extended to reconstruct multi-part sculptured objects. Making use of visually salient features along object occluding contours, contour segmentation is done. Occluding contours from multiple viewing points...[ Read more ]

In this thesis, we propose a 3D reconstruction method based on passive sensing techniques by integrating multiple visual cues. Intensity images from multiple viewpoints are taken as input data. Multiple visual cues such as volumetric intersection, occluding contours and stereo contribute together in the process of 3D reconstruction. Deformable superquadrics are chosen as geometric primitives for object representation. Physics-based approach is adopted to simulate the 3D reconstruction process, where the visual modules cooperate with each other. More importantly, the proposed method is extended to reconstruct multi-part sculptured objects. Making use of visually salient features along object occluding contours, contour segmentation is done. Occluding contours from multiple viewing points give the position and dimension estimate for each part. Ways of dealing with data missing due to occlusion occurring between constituent parts are also discussed.

The multi-part object reconstruction method is applied to both mannequin and real human body. By incorporating domain-dependent knowledge, a virtual measuring system is also developed to supply customerized measurements and features from the reconstructed objects. A complete system for 3D data capturing is developed during the research, specifically for human body reconstruction.