General heterogeneous sensor estimation fusion-system fusion method

This paper provides a summary of general weighted sensor estimate Mean Square Error (MSE) fusion and recently established Minimizing Euclidean Error Estimation (MEEE) fusion. Based on the MEEE setting, we propose a general heterogeneous sensor estimation fusion method. Unlike the previous estimation fusion method, the statistical correlations between sensor estimation errors are not needed, as well as, a quantitative joint function relationship between the multiple heterogeneous sensor estimates also is not needed in the new method. In practice, it is so hard to get the both information. This is why so difficult to fuse heterogeneous sensor data before. All we need are the sensor estimates and their error bounds. Obviously, a rough estimation error bound can be derived much easier than an accurate statistical correlation or a quantitative joint relationship between sensor estimate errors. Instead, a reasonable way to establish their organical connection between the multiple heterogeneous sensor estimates is to remodel all sensor estimates and their error bounds to be a group of measurement equations for the given estimation problem. While the MEEE for this remodelled system is derived, the heterogeneous sensor estimation fusion can be completed by using MEEE. Hence, we call it System Fusion Method (SFM) to differentiate from the most popular distributed weighted fusion method. Since the estimate of an algorithm can be reviewed as a measurement of a soft/nominal sensor, the proposed method can be used to fuse the results from multiple heterogeneous algorithms. It is easy to see that this method is quite general and independent of specific heterogeneous sensor data, and builded on solid theoretical basis of optimality since MEEE is founded on contemporary convex optimization and intersection fusion of estimation coverage sets of true value as given in [1]. Besides, we present some suggestions how to choose better error bounds and apply a multi-error-bound method to handle lack of the error bound knowledge.