Conclusion

Data fusion can only be effective if uncertainties in the information from the various sources are available, or estimated. It is therefore desirable that each element of the system provides a soft decision, i.e. a likelihood vector (a probability density over the possible values of the state), and if this is not the case, that at least the fusion nodes have information on the qualities of the sources. As soon as this minimum is reached, we see surprisingly good performance from fusion operators weighted by source quality.

A second success factor is to manage the informational redundancy inherent in this type of system, be it the correlation of inter-sensor errors or the uncontrolled duplication of data elements within the system. Without special precautions, disastrous decision biases are observed. Ad hoc methods can usually deal effectively with these problems....