Data Compression Methods For Multisensor Kalman Filter Fusion

In recent years, multisensor Kalman filter fusion have received significant attention for both military and non-military applications. But in fusion center the data size may be too large to be processed immediately. In 1976 [1], a data compression method for multisensor Kalman filter fusion was brought forward firstly. In that case, the data are compressed in the fusion center to improve the computational efficiency. But the measurement data have to be transformed to a common coordinate system. Obviously, this result is too limited to be applied. In this paper, two new methods are proposed to deal with the more common situations and their arithmetic equivalence to uncompressed processing is proved.

Alternate abstract:

多传感器的Kalman滤波融合是在许多军事和民用高科技部门中有重要应用的问题，在融合中心常常会因观测数据过大而不能实时处理。在文献[1]中首次提出对于多传感器的Kalman滤波融合，中心先将收到的观测数据进行压缩，以在不损失性能的情况下减少融合中心的运算量，但要求所有传感器的观测矩阵必须能够通过坐标变换转化为同一矩阵，即每个观测矩阵的行向量张成的子空间必须相同，才能保证用压缩后的数据仍能得到与无数据压缩时相同的最好性能。这一结果以后被许多文献引用并收入专著如[2,3]。显然文献[1]的条件过于限制，一般并不满足。当这一条件不满足，文献上一直未见观测数据如何压缩的结果。本文将针对当所有的观测矩阵不一定都在同一子空间内的一般情况进行讨论，并提出两种观测数据压缩的解决方案，并证明用压缩数据滤波性能没有损失。