An accurate and robust visual-inertial positioning method

Abstract

The human-machine integrated coordinate measurement is a promising coordinate measurement method with high flexibility and efficiency for the complex working environments. The cameras installed on the head-mounted measurement device (HMMD) achieves accurate global positioning by observing the uncoded LED landmarks, and then combines with the local measuring to obtain 3D coordinates. However, limited by the frame rate of the camera, the fast movements of the operator's head may cause landmark misidentification and visual positioning failure. In order to improve the robustness, a visual-inertial positioning method is proposed in this paper. An inertial measurement unit (IMU) is added to compensate for the deficiency of the visual positioning and enhance the dynamic performance. An adaptive extended Kalman filter (EKF), which adjusts the measurement noise covariance matrix based on the visual positioning uncertainty, is established to obtain the optimal state estimation. And an efficient initialization procedure is presented to implement the initial registration of uncoded landmarks based on the Normal Distribution Transform (NDT) algorithm and to determine the initial state of the IMU. Furthermore, the residual chi-square test is employed to detect false pose estimate in real time and to avoid positioning failure. The experiments demonstrate that the proposed method has high static positioning accuracy (0.681 mm) and high dynamic positioning robustness. The adaptive EKF realizes reliable landmark identification under fast movements and provides a higher accuracy than the common EKF.

Air-ground multi-agent system cooperative navigation based on factor graph optimization SLAM

“…To provide a comprehensive understanding of the algorithm design process, it is necessary to delve into an explanation of the EKF framework. The algorithmic procedure primarily comprises four sequential steps: initially, the extraction of feature points within each data frame; secondly, the initiation of the agent's pose and feature point information; subsequently, the prediction of the robot's pose information for the subsequent moment based on the system's state model; and finally, the update of the system's state estimate and covariance matrix through the application of the Kalman gain [18,19].…”

A GNSS/SINS fault detection and robust adaptive algorithm based on sliding average smooth bounded layer width

“…When the GNSS measurement data is good and the system is fault-free, the Gaussian distribution of the mean value of the innovation residual e k is considered to be zero. When the GNSS measurement data is abnormal and the system fails, the innovation residual e k does not conform to the Gaussian distribution with a mean of zero [27,28]. The hypothesis testing problem posed by making binary assumptions about the innovation residual can be expressed as follows:…”