中文摘要：

在传统的以惯性与视觉相对运动参数差为观测量的惯性／视觉组合导航模型中，为了得到简单的观测矩阵，惯性相对旋转四元数与平台失准角之间的函数关系往往通过近似处理得到。但当航向角较大时，该近似处理的误差很大，不可忽略。为解决该问题，文章提出了一种不需该近似处理的改进的观测量误差分析方法，并建立了改进的观测模型。该方法利用含误差的惯性姿态已知的特点，得到了更精确的惯性相对旋转四元数误差表达式，减少了Kalman滤波中惯性与视觉相对旋转差的状态模型预测误差。月面仿真和地面试验表明该改进方法相比传统方法可实现更高的导航精度，其中位置精度提高达68％。

英文摘要：

The INS/VNS integrated navigation is an important autonomous navigation method for lunar rovers. Traditional INS/VNS integrated navigation methods which utilize the relative motion differences between the inertial and the vision as measurements, usually make an approximation between the inertial attitude errors and the misalignment angle to simplify the relative rotation measurement model. However, the approximation error can exceed the misalignment angle when the head is large, which cannot be ignored. To solve this problem, a new measurement error analysis method was presented which can avoid the approximation in traditional method, and build the corresponding improved measurement model. The known inertial attitude which contains error utilized to get more accurate error expression of relative inertial rotation quaternion, and reduce the forecast error of relative rotation difference between the inertial and the vision. Both of the Moon-based simulation and the Earth-based experiment demonstrate that the presented method is effective and can achieve higher accuracies of positions and attitudes than the traditional one, in which the position accuracy has improved 68%from 0. 456% to 0. 143% in the Earth-based experiment.