DYNAMIC VIEW MORPHING

Dynamic view morphing deals with the scenes containing moving objects in presence of camera motion. In this project, we use a novel technique for synthesizing dynamic scene between two images without use of 3D model. The scene contains rigid or non-rigid objects, each of them can move in any orientation or direction. Our method can generate a series of continuous and physically accurate intermediate views from only two reference images without 3D knowledge. The procedure consists of three steps: segmentation, morphing and postwarping. Given a boundary connection constraint, the source and target scenes are segmented into several layers to morph. Based on the decomposition of affine transformation between corresponding points, we uniquely determine a physically correct path for postwarping by the least distortion method.

In this project, we successfully generalized dynamic scene synthesis from simple rotation case to dynamic scene containing non-rigid objects. Our method can handle these cases mixed with dynamic rigid or non-rigid objects, including complicated object such as humans. In our approach, if the non-rigid object can be segmented into several rigid parts, we can use the boundary information to connect adjacent parts to morph two reference views, and generate a series of continuous and physically accurate intermediate views without the 3D knowledge.

Results:

Dynamic view morphing for human action

Two reference views: