Feature Extraction and Classification of Natural Landmarks for Navigation of Mobile Robots

초록

In order for mobile robots to explore an uncertain environment, nearly all approaches are used like vision-based navigation methods based on tracking feature points or landmarks in images. In most mobile robot implementations, the robot’s landmark recognition system is provided with domain-specific recognizable landmarks such as lane boundaries, ceiling light, bar codes, door edges, etc. In this paper, however, we consider the problem of exploring an unfamiliar environment in search of recognizable objects or visual landmarks. The visual landmarks are recognizable objects or patterns that can be very salient and readily distinctive. And we use these two important characteristics. The salient characteristic means that the landmark should readily “pop out” from the background by some detection mechanism and the distinctiveness means that the landmark is unlikely to be confused during recognition. In order to extract and recognize them automatically, we construct a feature map that records the set of features continually during a learning phase. The map contains photometric and geometric information of the features as well as their metric information, and also contains probability of its existence in the environment. We demonstrate the landmark selection process and the feature map building procedure, and present that our method can be used for mobile robot navigation.