Abstract:
This thesis presents a method for direct ray tracing of point sampled surfaces. This allows to render high quality images using unstructured point-sampled data. An oriented disc is placed at each point and rays are tested for intersection with these discs. In case of intersection, all points within a xed distance of the ray are used to interpolate the position, normal and any other attributes. Ray-Disc intersection tests are accelerated using a hierarchical data structure namely an octree. This provides a computation complexity of O(n log n). Also searching for intersected octree nodes around ray is much more faster than scanning whole point data. All of the intersected discs can be located inside the nodes which are intersected by the ray. For data without point normals, required normals are calculated from point data by tting a tangent plane to each point. An improvement is proposed for adapting disc radius to the local point density by determining k nearest neigbours. Using variable disc radius lls empty holes within the point cloud without loss of detail. Variable disc radius approach is also used in normal estimation process where the k nearest neighbours are used in calculations instead of neighbours within a xed radius.
