3D eye position based interaction within hierarchically represented images
Our hierarchical image representation method is composed of multiple layers. On the lowest level, we will recover the 3D extrinsic camera parameter of images which will build the foundation of our system. Upper layers of our structure are separated with clustering algorithm, in which the feature space established within the lowest layer consists of the camera's 3D position and orientation. In this section, we will explain how we extract the 3D geometry from images, establish a relationship of 3D geometries from multiple images and adequate classification methods of images at the upper layers. Figure 4 shows the flow chart for a hierarchical representation method of an image set. With SIFT, RANSAC, the epipolar geometry and 3D camera position of a set of online images is estimated as shown in Figure 4. The lowest layer of our hierarchical structure is constructed using the 3D geometry of the images. In a next step, we describe the clustering of images based on the distance of each 3D camera position within the world coordinate system.