Ray tracing is a graphical technique that provides realistic simulation of light sources and complex lighting effects within three-dimensional scenes, but it is a time-consuming process that requires a tremendous amount of compute power. In order to reduce the number of calculations required to render an image, many different algorithms and techniques have been developed. One such development is the use of tree-like data structures to partition space for quick traversal when finding intersection points between rays and primitives. Even with this technique, ray-primitive intersection for large datasets is still the bottleneck for ray tracing. This thesis proposes the use of a specific spatial data structure, the K-D tree, for faster ray casting of primary rays and enables a ray-triangle culling technique that compliments view frustum and backface culling. The proposed method traverses the entire tree structure to mark nodes to be inactive if it is outside of the view frustum and skipped if the triangle is a backface. In addition, a ray frustum is calculated to test the spatial coherency of the primary ray. The combination of these optimizations reduces the average number of intersection tests per ray from 98% to 99%, depending on the data size.

Raytracing is a basic technique for displaying 3D objects. The goal of this thesis is to demonstrate the possibility of implementing raytracer using a programmable GPU. The algorithm and its modified version, implemented using "C for CUDA" language, are described. The raytracer is focused on displaying dynamic scenes. For this purpose the KD tree structure, bounding volume hierarchies and PBO transfer are used. To achieve realistic output, photon mapping was implemented.

This thesis deals with the method of distributed ray tracing focusing on optimalization of this method. The method uses simulation of some attributes of light by distributing rays of lights and it produces high quality and partly realistic images. The price for realitic effects is the high computational complexity of the method. The thesis analysis the theory connected with these aspects. A large part describes optimalizations of this method, i.e. searching for the nearest triangle intersection using kd-trees, quasi random sampling with faster convergence, the use of SSE instruction set and fast ray - triangle intersection. These optimalizations brought a noticable speed - up. The thesis includes description of implementation of these techniques. The implementation itself emphasises the practical usability including generating some advanced animations and universal description of objects.