Object recognition to support indoor robot navigation

This work presents how we can apply feature descriptor and classifier methods to support indoor navigation for robots. Our aim is to recognize objects (such as doors, elevators/lifts, chairs, fire extinguishers, sockets) in man-built environments where our robot can move. Having recognized the objects, information as to where each object can be found is transmitted to the mapping module for recording. This way the accuracy of map building and navigation can be improved significantly. To solve this problem we use feature detector and descriptor methods, such as SIFT or SURF, to narrow the scores before we apply a support vector machine. This is a classification method that, following a suitable training pattern, sorts the input images into the learned classes. Our solution follows out a kind of classification based on a `Bag-of-Words´ feature vector with the help of a support vector machine. Based on the result of this classification we do the SURF localization process with the help of carefully chosen model images. Combining these two algorithms and the models stored in a database, the recognition is manageable with success.