|M.Sc Thesis||Department of Computer Science|
|Supervisors:||Assoc. Prof. Barequet Gill|
|Prof. Gotsman Chaim Craig|
Many applications, ranging from visualization applications (such as architectural walkthroughs) to robotic applications (such as surveillance), could benefit from an automatic camera trajectory planner. This thesis deals with that problem. We have automated the process of inspecting the outside of a simple two-dimensional polygon, given a few user parameters. Our aim is to find a closed path for the camera's position and direction, so that the shape is filmed in a visually-pleasing way. Creating such a trajectory manually is a time-consuming task for an unexperienced animator, architect or 3D graphics artist. The focus of this work is cinematographic, since we are trying to find a trajectory for a camera which produces a given desired effect.
Our algorithm preprocesses the polygon using visibility-graph-like concepts, and creates a visibility data structure for each polygon edge. These structures contain the optimal camera location and view direction per polygon edge. From these structures, "good" camera zones are computed. Given an ordered list of simple polygons (the camera zones), each two consecutive polygons with a nonempty intersection, we find a path lying entirely in the union of these polygons, passing through each in turn, and minimizing a given cost function. Natural cubic splines (an interpolation scheme) are then used to create a closed camera trajectory that passes solely inside the zones. An iterative process refines the trajectory by minimizing a cost function until it converges to a good result. The three-dimensional path that we obtain is smooth and the visualization of the polygon when traveling the path is visually pleasing.