Autonomous cars are mobile robots that navigate through highly unpredictable environments such as city traffic. Obstacle detection is one of the fundamental tasks they have to be capable of. This work describes the implementation of a stereo vision based real-time obstacle detection system for an autonomous model car on the scale 1:10 built for the annual Carolo-Cup contest. First, a modern stereo camera is introduced to the existing platform. The provided depth information serves as the basis for an obstacle detection algorithms that is adapted to the model car's reduced environment. White cardboard boxes placed on a circular track have to be detected as obstacles to enable the car to navigate around them. The algorithm detects elevations above the ground as well as the distance and width of these obstacles, and models them as a bird's eye view of the car's surroundings. A special focus is placed on the obstacle detection system's runtime, as the cars main processor only provides limited computational power. The algorithms makes heavy use of integral images and specialized image pyramids to enable execution with 20 Hz and more.