Closed-form metric velocity and landmark distance determination utilizing monocular camera images and IMU data in the presence of gravity

Abstract

In this paper we present the enhancement of an existing closed-form solution for metric velocity determination utilizing monocular images in combination with accelerometer and gyroscope measurements. While the original version of this algorithm depends on external attitude information for gravity compensation, our solution allows for gravity compensation through the addition of magnetometer measurements. The proposed algorithm results in a linear system of equations which is solvable for accelerated trajectories of an agent carrying the aforementioned sensors. If this condition is met the solution directly provides scaled metric information about the velocity, the distance to landmarks and the direction of gravity in the agent's frame. The authors propose consideration of this algorithm for the initialization of monocular vision navigation filters.

Metric velocity, landmark distance and attitude estimation utilizing monocular camera images and gravity biased IMU data

“…Recently an augmentation of the algorithm from [14] was presented in [1], coping with gravity biased IMU data by incorporating one additional landmark observation and magnetometer measurements.…”

“…In this paper we propose the combination of the closed-form solution presented in [1] and the iterative filter approach presented in [2]. The resulting filter framework for metric velocity, landmark distance and attitude estimation does not require any prior information or assumptions regarding the desired estimates or the agent's environment.…”

“…To ensure a highly consistent initialization, it is desirable to not apply any assumptions about the state and covariance, but to calculate them from measurements. Therefore the closed-form solution from [1] will be applied and briefly outlined. Furthermore an approach for the calculation of the initial covariance based on this closed-form solution will be presented.…”

“…The framework combines two newly developed algorithms for metric velocity and landmark distance determination. The first algorithm is a closed-form solution [1] which is applied for filter initialization. The second one is an alternative Extended Kalman Filter (EKF) formulation [2], which directly yields in scaled state estimates and features linear output equations.…”

Navigation and locomotion of a low-cost Automated Guided Cart

“…Although AGCs are finding ever-expanding use in the South African manufacturing industry, there is little local development besides that which takes place at academic institutions. Furthermore current research into AGVs is focused on finding new methods for navigation, mobilisation [1] and odometry [2], not focusing on cost. While important for the future development of AGVs; these systems are costly and difficult to implement into a basic AGC.…”