Subject :

Augmented reality is a technique allowing the insertion of virtual objects in a real environment. Applications are numerous : reconstitution of lost monuments, virtual guide for museum visit, user assistance for manipulation of complex machines. The augmented reality process has two steps : the first is computing the virtual object and the second is the insertion of this object in his environment. For this thesis, we will be interested in this second point. To insert an object, we have to know geometry and lighting of the environment. Therefore R3AM team of LIRIS laboratory propose a thesis on real time acquisition of environment geometry and photometry for augmented reality. Ideally, this environment could be stored in high dynamic range (HDR) images taken from many viewpoints, eventually combined with depth information.

Objectives of the thesis :

- The first objective is to acquire efficiently and easily an HDR image. Contrary to classical images, HDR images store for each pixel a physical value of lighting that is not an integer limited between 0 and 255. So, they are able to contain all different lightings of the scene. This technique will be used to acquire lighting characteristics in one particular point of the environment. Actually, acquisition of these images is only possible by combining classical images having different exposure. A state of the art of existing methods and speed improving methods has been done in master thesis. We have now to use these methods in the complete augmented reality framework developed in our team.

- The second objective is to obtain environment geometry and camera position from multiple view of the scene. Many methods exists, so it will be necessary to make the state of the art, test some of them and maybe propose a method designed for real time application. We could first recover disparities (i.e. depth map) of images with one of the classical approach of this problem : estimation of depth from camera focus, computing disparity from a pair of stereoscopic images... The depth from focus method is in evaluation in a master thesis. This work will be continued in the thesis and other methods will follow.

- The third objective is to efficiently use these environment acquisitions in order to synthesize images. We could try an image-based approach and estimate how it is possible to mix HDR images. Occlusions will be manage with depth map and we can also imagine an object relighting by the environment. This part is extremely complex and will be developed in collaboration with Etienne Stanowski, who is already working as Phd student in our team.