Development Challenges and Performance Analysis of Drone Visual/Inertial SLAM in a Global Reference System

Abstract

this paper presents a navigation system design for airborne drone-based applications. The proposed operation principle includes visual/inertial tightly coupled integration based on an extended SLAM approach. In addition to the opportunity-based features that are used by similar to the standard SLAM solutions, the proposed approach also utilizes internal landmarks that are created by the algorithm itself. This method allows reducing the drift that is typical in SLAM -based solutions in addition to reducing the algorithm complexity. Adding a small number of global landmarks helps eliminate the remaining drift. The design goal is to use global coordinates during the entire operational cycle. For a 200m altitude flight, with 2 global landmarks, partially available, the system converges fast to high accuracy with respect to the global reference, without any SLAM drift. While for lower altitudes (100 m), only near-tactical grade IMU provides a very low SLAM drift solution. The significance of the work lies in showing the complete design of a global-coordinate drone SLAM system with the fusion of inertial sensors, taking into account tradeoffs on trajectory profile, sensor grades and global landmarks availability.