Efficient object detection
A long-term goal of computer vision is to interpret visual sceneries, in other words, to let computer perceive like humans do. To this end, computers have to localise and categorise objects in images. This fundamental task is called object detection and is the subject of this work. The difficulty of this task lies in the vast number of possible object locations and the strong variations of an object's appearance. This dissertation focuses on efficient localisation; we analyse and discuss several strategies. These investigations lead to a novel branch&rank algorithm that detects with often less than 100 ranking operations. This allows for rich appearance models (like non-linear SVMs) which eventually improves overall performance.