Disparity as a separate measurement in monocular SLAM

Abstract

In this paper, we investigate the effects of including disparity as an explicit measurement (not just for initialization) in addition to the projective camera measurements, in conventional monocular/bearing-only SLAM in a 2-D world. We conduct an observability analysis for a 1.5-D scenario which theoretically shows that adding disparity measurements influences the observability rank condition, and so disparity can be regarded as a valid measurement. As disparity is dependent on the camera's location one time step earlier, we rederive the EKF equations for measurement models dependent on the previous state. It is shown that this modification results in an increase in the measurement noise covariance. We also discuss the implications of the white noise assumption in EKF being violated with the inclusion of disparity measurements. In the end, we compare the results of a simulated monocular SLAM scenario in a 2-D world with and without disparity measurements. Here we use the magnitude of the eigenvalues of the simulated state covariance matrix as a metric for observability [1]. Our results suggest that including disparity measurements in monocular SLAM improves observability both in the map and pose variables. Hence, this paper is an invitation to consider disparity measurements explicitly, in monocular SLAM.