Comparative Study between EKF, SVSF, Combined SVSF-EKF, and ASVSF Approaches based Scale Estimation of Monocular SLAM

Abstract

: This paper presents a comparative study of scale recovering in monocular simultaneous localization and mapping (Mono-SLAM) by adopting and adapting four estimators into a multi-rate fusion mechanism and considering the scale as an element of the state vector. These estimators are: extended Kalman filter (EKF), smooth variable structure filter (SVSF), combined SVSF-EKF, and particularly adaptive smooth variable structure filter (ASVSF). The use of the ASVSF estimator represents the novelty of this paper because it provides a robust estimation of the trajectory scale as well as the covariance matrix at each iteration. This later represents the estimation incertitude. A second sensor is involved (inertial measurement unit (IMU)) as a reference to align the up to scale trajectory provided by the Mono-SLAM box. The designed system allows finding the scale factor with a rate not further than the IMU frequency and avoids complex synchronization. In order to outline the limitation of each estimator used for scale recovering, a deep analysis of the proposed approaches in terms of robustness, stability, accuracy, and real-time constraint was carried out.