A practical approach for EKF-SLAM in an indoor environment: Fusing ultrasonic sensors and stereo camera

Improving the practical capability of SLAM requires effective sensor fusion to cope with the large uncertainties from the sensors and environment. Fusing ultrasonic and vision sensors possesses advantages of both economical efficiency and complementary cooperation. In particular, it can resolve the false data association and divergence problem of an ultrasonic sensor-only algorithm and overcome both the low frequency of SLAM update caused by the computational burden and the weakness to illumination changes of a vision sensor-only algorithm. In this paper, we propose a VR-SLAM (Vision and Range sensor-SLAM) algorithm to combine ultrasonic sensors and stereo camera very effectively. It consists of two schemes: (1) extracting robust point and line features from sonar data and (2) recognizing planar visual objects using a multi-scale Harris corner detector and its SIFT descriptor from a pre-constructed object database. We show that fusing these schemes through EKF-SLAM frameworks can achieve correct data association via the object recognition and high frequency update via the sonar features. The performance of the proposed algorithm was verified by experiments in various real indoor environments.