To analyze scenery obstacles in robotics applications depth information is very valuable. Stereo vision is a powerful way to extract dense range information out of two camera images. In order to unload the CPU the intensive computation can be moved to GPU, taking advantage of the parallel processing capabilities of todays consumer level graphics hardware. This work shows how an efficient implementation on the GPU can be realized utilizing the NVIDIA Cuda framework.
In this thesis we apply powerful mathematical tools such as interval arithmetic for applications in computational geometry, visualization and computer graphics, leading to robust, general and efficient algorithms. We present a completely novel approach for computing the arrangement of arbitrary implicit planar curves and perform ray casting of arbitrary implicit functions by jointly achieving, for the first time, robustness, efficiency and flexibility. Indeed we are able to render even the most difficult implicits in real-time with guaranteed topology and at high resolution. We use subdivision and interval arithmetic as key-ingredients to guarantee robustness. The presented framework is also well-suited for applications to large and unstructured data sets due to the inherent adaptivity of the techniques that are used. We also approach the topic of tensors by collaborating with mechanical engineers on comparative tensor visualization and provide them with helpful visualization paradigms to interpret the data.