Résumé : This paper presents a new way to insert virtual objects into real environments. It proposes an image based method to determine light interactions between the different parts of the scene. Unlike most previous works, our tech- nique does not need any manual modeling: data are extracted from multiple 2.5D high dynamic range captures of the scene. These images are obtained via a calibrated time of flight camera coupled with an ordinary webcam. Visibility infor- mation and light transport are stored using a voxel grid. Us- ing these data, we create an new image with a seamlessly inserted virtual object. We demonstrate the efficiency of our method by inserting a virtual object first in a virtual environ- ment and next in a real environment.