3D Reconstruction of Dynamic Scenes with Multiple Handheld Cameras

摘要: Accurate dense 3D reconstruction of dynamic scenes from natural images is still very challenging. Most previous methods rely on a large number fixed cameras to obtain good results. Some these further require separation static and points, which are usually restricted with known background. We propose novel depth estimation method can automatically recover accurate consistent maps the synchronized video sequences taken by few handheld cameras. Unlike camera arrays, our data capturing setup much more flexible easier use. Our algorithm simultaneously solves bilayer segmentation in unified energy minimization framework, combines different spatio-temporal constraints for effective optimization points. A variety examples demonstrate effectiveness proposed framework.