误差状态卡尔曼滤波的视觉惯性自适应融合定位方法研究

Vision Inertia Adaptive Fusion Method for Attitude Determination Based on Error State Kalman Filter

头盔瞄准具(HMS)是近年来新一代战斗机飞行员的辅助瞄准设备,能够帮助飞行员增强战场态势感知能力,对敌方目标进行快速、精准打击。其能正常工作的关键是获取飞行员头部相对于运动飞机的姿态参数。本文结合头盔瞄准具这一应用场景研究了视觉组合姿态测量关键技术。视觉惯性组合定位能够实现目标位姿测量方法的优势互补,而由于标称噪声矩阵无法绝对准确预测,融合算法的鲁棒性、精度有待进一步提升。针对这一问题,本文提出一种误差状态卡尔曼滤波框架下基于变分贝叶斯推断的视觉惯性自适应融合方法。首先,对于过程噪声使用逆威沙特(Wishart)分布进行建模,之后通过引入隐变量分解一步预测协方差,并结合变分贝叶斯推断实现了对过程噪声协方差矩阵的在线估计。试验证明,在复杂运动及标称噪声协方差矩阵偏移较大的测量条件下,所提位姿测量算法具有较高的精度与鲁棒性,能够完成对靶标的快速、高精度跟踪。

Helmet Mounted Sights(HMS)are auxiliary sighting equipment for new generation fighter pilots in recent years.They can help pilots enhance battlefield situational awareness and conduct rapid and precise strikes against enemy targets.The key to its normal operation is to obtain the attitude parameter of the pilot’s head relative to the moving aircraft helmet-mounted sight.This paper investigates the key techniques for visual fusion and posture measurement in the context of helmet mounted sight.The visual inertial fusion method can realize the complementary advantages of these two target position measurement methods.However,the robustness and accuracy of the fusion algorithm need to be further improved,because the nominal noise matrix cannot be predicted absolutely and accurately.To address this problem,this paper proposes a visual inertial adaptive fusion method based on variational Bayesian inference in the error-state Kalman filter framework.First,the process noise is modeled using the inverse Wishart distribution.Then,the covariance is predicted in one step by introducing a latent variable,and the online estimation of the process noise covariance matrix is achieved by combining the variational Bayesian inference.Experimental findings unequivocally demonstrate that the proposed pose measurement algorithm exhibits remarkable accuracy and robustness in the face of complex motion and substantial deviations in the nominal noise covariance matrix.The proposed algorithm can complete fast and high-precision tracking of the target.