Improving Data Association in Robot SLAM with Monocular Vision

In the paper, an algorithm is proposed for improving the data association in robot visual Simultaneous Localization and Mapping (SLAM). The detection of speeded-up robust feature (SURF) is employed in the algorithm to provide a robust description for image features as well as a better representation of landmarks in the map of a visual SLAM system. Meanwhile, a likelihood-based tracking window and a nearest-neighbor (NN) method are utilized to match the high-dimensional data sets created for SURF. Experiments are carried out on a hand-held camera to verify the performances of the proposed algorithm for dealing with the data association problem in robot visual SLAM. The results show that the integration of the SURF features, the tracking window and the NN method is efficient in reducing the computational time and increasing the rate of successful feature matching.