Visualisation of videoconference image sequences using VRML 2.0

In this paper a procedure for visualisation of videoconference image sequences using Virtual Reality Modeling Language (VRML) 2.0 is described. First, image sequence analysis is performed in order to estimate the shape and motion parameters of the person talking in front of the camera. For this purpose, we propose the K-Means with connectivity constraint algorithm general segmentation algorithm combining information of various types such as colour and motion. The algorithm is applied "hierarchically" in the image sequence and it is first used to separate the background from the foreground object and then to further segment the foreground object into the head and shoulders regions. Based on the above information, the 3D shape parameters are estimated for each sequence and a 3D model is automatically adapted. The rigid 3D motion is estimated next for each sub-object. Finally a VRML file is created containing all the above estimated information and can be viewed using any VRML 2.0 compliant browser.