An Enhanced Fusion Strategy for Reliable Attitude Measurement Utilizing Vision and Inertial Sensors

Abstract

In this paper, we present a radial basis function (RBF) and cubature Kalman filter (CKF) based enhanced fusion strategy for vision and inertial integrated attitude measurement for sampling frequency discrepancy and divergence. First, the multi-frequency problem of the integrated system and the reason for attitude divergence are analyzed. Second, the filter equation and attitude differential equation are constructed to calculate attitudes separately in time series when visual and inertial data are available or when there are only inertial data. Third, attitude errors between inertial and vision are sent to the input layer of RBF for training. After this, through the activation function of the hidden layer, the errors are transferred to the output layer for weighting the sums, and the training model is established. To overcome the problem of divergence inherent in a multi-frequency system, the well-trained RBF, which can output the attitude errors, is utilized to compensate the attitudes calculated by pure inertial data. Finally, semi-physical simulation experiments under different scenarios are performed to validate the effectiveness and superiority of the proposed scheme in accurate attitude measurements and enhanced anti-divergence capability.