Monocular visual inertial navigation for multirotor under thrust command

Abstract

Visual inertial Navigation can be used for state estimation in a situation in which acquisition of GPS measurement is limited. In particular, since its low structural complexity and low power consumption, filter-based monocular visual inertial navigation is suitable for multirotor which is sensitive to size and power consumption. In this paper, a tightly coupled monocular visual inertial navigation that improves accuracy by adding multirotor dynamics and thrust command to filter measurement update process is proposed. Proposed model-aided visual inertial navigation is based on the ESKF(error state Kalman filter). State propagation of the filter is performed using angular rate and acceleration measurements obtained at every cycle. Measurement update of the filter is performed using thrust command and KAZE features extracted from the images obtained from monocular camera. The superiority of the proposed model-aided visual inertial navigation is demonstrated by comparing it with the existing filter based monocular visual inertial navigation that does not use the multirotor dynamics and thrust command.