Z˜(inf) - Monocular Localization Algorithm with Uncertainty Analysis for Outdoor Applications

Localization is an essential task for a meaningful application of a robotic system. The knowledge about its 6DoF pose represented as a 3D position and the orientation in space allows a system to execute tasks defined in the 3D workspace and to prevent collisions using the knowledge about obstacles from a model of the environment. A system needs to be capable of estimating its absolute location in the world and to track the changes in the position during its motion. We speak in this context of global and relative localization capabilities. The localization task performs a registration of the sensor perception to a reference. The two above localization alternatives differ in the way how this reference is defined. The global localization requires an a-priori knowledge about the environment stored usually as a geometric model or as an image database that needs to be registered to the current sensory perception, see (Thrun, 2002) for a comprehensive survey. Since early work in the 1970’s, such as SRI’s Shakey (Nilsson, 1984) and Moravec’s Cart (Moravec, 1983), there have been great strides in the development of vision-based navigation methods for mobile robots operating both indoors and outdoors. Much of the efficiency and robustness of the recent systems can be attributed to the use of special purpose architectures and algorithms that are tailored to exploit domain specific image cues. For example, road