EM-EKF based visual SLAM for simple robot localization

This paper presents a novel SLAM method based on the filter that fuses EM algorithm in EKF. Due to the visual SLAM mostly depends on the sensor information that is hard to be obtained in high-level accuracy, the proposed filter is designed to deal with this problem since it can estimate the unknown parameter from the known information in every frame. Realtime experimental results also prove the advantages of SLAM method based on the proposed filter. Compared to the regular EKF, SLAM based on EM-EKF is suggested to have up to 60 percent improvement in the accuracy. It also shows the advantage in the convergence speed and the stability of the system.