| dc.description.abstract | When simulating a moving observer or sensor in a 3D scene which contains a large number of objects, only the objects lying in the field of view or of interaction, named the local data base, have to be taken into account in order to decrease, at each step, computation time necessary for either image generation or various processings. This paper presents an algorithm which achieves the dynamic management of the local data base, that is to say which manages the set of objects which enter or leave the field of view of an observer (camera, sensor) during its displacement. This method consists, broadly, of the following two steps: a binary space partitioning of the 3D space is performed off-line from the object bounding boxes by means of planes perpendicular to the X, Y and Z axes and a subregion adjacency graph is created; the dynamic management of the local data base is achieved on-line by modelling the bounded bearing volume of the sensor by a cube of R "radius" (half-edge), and by using the adjacency graph, and inclusion and intersection criteria in order to exploit spatial and temporal coherences between each displacement. When the scene data base is very large, the dynamic management of the RAM memory can be done simultaneously by using this method reasoning from the bounding boxes of disjoined sub - scenes. | en_US |
