Omnidirectional Image Matching for Vision-Based Robot Localization

摘要

The new approach to utilize information from an omnidirectional image is presented in this paper. We describe the integration between the image matching and the Monte-Carlo localization that can be implemented in a large indoor environment. The robot is able to localize with reasonable accuracy despite no metric measurements from the image. Furthermore, the algorithm can recover quickly from localizing a robot at a wrong place