Research on binocular vision absolute localization method for indoor robots based on natural landmarks

Abstract

A method is put forward to extract feature points in the indoor ceiling corner as natural landmarks by using binocular visual system and to go further to achieve absolute localization for robot. Firstly, this method puts forward a kind of double threshold Features from Accelerated Segment Test (FAST) - Scale Invariant Feature Transform (SIFT) matching algorithm, in which FAST algorithm is applied to extract the feature points in the ceiling corner to reduce the number of feature points effectively. For each feature point, SIFT descriptor and matching algorithm is remained so that the advantages of the SIFT algorithm is inherited. And then the reasonable sampling rules are introduced to create natural landmark database, the SIFT descriptors of feature points in the ceiling corner and their 3-D world coordinates are saved into natural landmark database in offline mode, When robot walks, the feature points are identified by matching left image with all SIFT descriptors in database and then their world coordinates are obtained. In the meantime, the distances between the feature points and robot are measured by matching left image with the right one. Finally, the world coordinates of the robot are computed according to the principle of triangular localization. To demonstrate the effectiveness of a series of method proposed in this paper, we do the corresponding physical experiments. Experimental results show that indoor robot absolute localization method proposed in this paper has a good performance in real-time, reliability and accuracy.