Abstract

As the capabilities of video standards and receiver hardware are increasing towards integrated 3d animations, generating realistic content is now becoming a limiting factor. In this paper we present a new technique of generating 3d content from reality, i.e. from video sequences acquired with normal TV cameras. The major aim is to provide the TV viewer with animated 3d reconstructions of athletic events in MPEG-4 over Digital Video Broadcast (DVB), which allows for an immersive experience via free navigation and interaction on the receiver side. As intervention in the actual scene, e.g. by markers, is often prohibited, markerless computer vision techniques are used on the images from normal broadcasting cameras for the accurate estimation of an athlete’s movements. The paper focuses on the key components for the realistic reconstruction of 3d geometric features, which are the calibration of moving TV cameras and the modelling of the moving athlete in its environment. 1