Real World Data Fusion

Fusion of data from multiple sensors has the promise of substantial improvement in system performance for many important applications. However, there are several practical issues that must be addressed to achieve such improvement: (1) residual bias errors between sensors; (2) dense multiple target environments; (3) unresolved data; (4) errors in data association between sensors; (5) sensor errors that are not fixed in time or space but which are not white noise either. We describe state-of-the-art algorithms that attempt to mitigate such problems. We show simple back-of-theenvelope formulas which quantify the situation, as well as one well known formula that is extremely pessimistic.