|M.Sc Student||Clarke Anna|
|Subject||Analytical Model for Spatial Perception from Dynamic|
|Department||Department of Aerospace Engineering||Supervisor||Professor Arthur Grunwald|
This research belongs to the area of advanced pictorial displays for aerospace applications. The focus is on displays for presenting dynamic multiple-object scenes. In situations requiring maneuvering in space and time, the perspective format is the most natural and appropriate presentation. However, effective design and optimization of perspective displays, requires a thorough understanding on how human observers reconstruct three-dimensional (3-D) visual scenes from dynamic images. Our approach for achieving insight into human spatial reconstruction is based on developing a mathematical formulation of the various interacting elements. This work presents an analytical model for 3-D structure and motion perception from perspective multiple-object image sequences. It is assumed that the observer perceives the direction of the lines-of-sight (LOS) to conspicuous points in the visual field, as well as the LOS rates. Inaccuracies in the perception are modeled by adding random observation noise to the LOS directions and their rates. The analytical model is based on the following four main elements: (1) A static spatial perception model, which uses a-priori knowledge of the objects in the scene in order to achieve reconstruction from a single image; (2) A motion perception model, which performs reconstruction from two sequential images, and is based on the rigidity assumption of the viewed object; (3) An object detection model, which uses three subsequent images to segment the scene into independently moving rigid objects; (4) A dynamic motion model, which constitutes a framework for assembling the three other perceptual elements. Since this last model is to provide the consistency of the estimation results over time, it was found to be a key factor in the reconstruction process. It was concluded that visual reconstruction of dynamic scenes involves matching information coming from three basic sources: (1) Perspective geometry and familiarity with the shape of the objects; (2) Observer’s motion as detected from the differences between successive views; (3) Correlation in time of the parameters, which describe the 3-D scene. The novelty of our work is the above multi-element reconstruction scheme, which allows to achieve a clear mathematical formulation of the stages and nuances of the reconstruction process, as well as of the observers’ inherent errors and limitations. It has been shown by means of numerical simulations, how the model can be used as a useful tool for the design and optimization of dynamic perspective displays.