Image Synthesis by Walker Replacement Based on Visual Motion Tracking

Abstract

One image synthesis technique involves replacing one object in an image sequence with another object. A significant problem in such an approach is how to locate objects in created images in a consistent manner considering occlusion among objects. For example, if a person in a sequence of images can be replaced by another person, it is easy to keep a reasonable positional relationship between the objects. This paper proposes a method for substituting a walker in a video stream with a 3-D CG model of another person. It is assumed that the image sequence should be taken by a fixed camera and the walker in the image sequence should walk straight. Motion parameters of the walker in the image sequence are obtained by tracking the walker using a stepwise 3-D model matching technique. This method can reduce the cost of obtaining the motion parameters of a human 3-D model and can detect self-occluding parts. The 3-D CG model of a person is constructed by measuring the person, regarding the person as an articulated object. Image synthesis can be achieved by replacing the extracted human regions with another image (or CG model). The experimental results show that the proposed method can achieve a walker replacer.