Image-based Analysis for Model-based Tracking

Large display environments allow to support collective visualization of Virtual Reality applications. As interaction in those environments can be troublesome, we considered the integration of gestural information into corresponding interaction techniques. To get rid of trackers and of some of the typical devices weaknesses, we investigated the use of video tracking. The realtime system we proposed combines a full use of graphics hardware and a parametric 3D model. 2D images are synthesized and compared to input frames. The latest improvement to our system lies in the collaboration between two traditionally distinct approaches. Indeed, model-based and image-based algorithms are in general concurrent. However, image-based preprocessing of input video frames enables us to bypass some of our initial limitations. Image moments permit to speed up hand position extraction, while image differencing cancels the minimization of some degrees of freedom.