Data fusion

§01What is data fusion?

Hall and Llinas’s widely cited tutorial in Proceedings of the IEEE frames multisensor data fusion as a set of applications, process models and techniques for combining data from multiple sensors.¹Source 1 · D. L. Hall and J. Llinas, Proceedings of the IEEE, 1997An introduction to multisensor data fusionieeexplore.ieee.org A widely used process model comes from the US Joint Directors of Laboratories (JDL). It separates the work into levels. The first covers association, correlation, estimation and identification in the data domain. Higher levels add knowledge-based reasoning about situations and their impact, then process monitoring and optimization, and human–computer interaction.²Source 2 · Ge Wang, BioMedical Engineering OnLine (PMC), 2005Review of “Mathematical Techniques in Multisensor Data Fusion” by Hall and McMullenpmc.ncbi.nlm.nih.gov

§02How does data fusion work in practice?

1. 01
   Step 1:

   Align

   Put every observation on a common clock, coordinate system and unit set, and keep its source and quality indicators.

2. 02
   Step 2:

   Associate

   Decide which observations belong to the same entity, using shared keys (24-bit address, MMSI) where they exist and spatial-temporal gating where they do not.

3. 03
   Step 3:

   Estimate

   Combine associated observations into one state (position, velocity, identity), weighting each by its reported or learned accuracy.

4. 04
   Step 4:

   Assess

   Flag disagreements between sources, score confidence, and route anomalies to a human.

§03Example: one aircraft, three sources

An aircraft broadcasts ADS-B with a low NACp because of GNSS interference. A multilateration network computes its position independently, and a Mode S reply supplies altitude. A fusion step that weights by quality leans on MLAT for position while the ADS-B accuracy is low, keeps ADS-B for callsign and velocity, and records that the two disagreed. This is the multi-sensor tracking idea that EUROCONTROL’s ARTAS applies operationally across radar, Mode S, WAM and ADS-B.³Source 3 · EUROCONTROLAir traffic management surveillance tracker and server (ARTAS)eurocontrol.int The integration of WAM into a multi-sensor tracker was also studied in EUROCONTROL’s WAM assessment.⁴Source 4 · National Aerospace Laboratory NLR for EUROCONTROL, 2005Wide Area Multilateration: Report on EATMP TRS 131/04eurocontrol.int

§04Common misconceptions

• “Fusion is just a join.” A join assumes the keys are right. Fusion has to decide whether they are, and what to do when sources disagree.
• “More sources always help.” An unreliable source with no quality indicator can drag a good estimate off. Weight by quality, or leave it out.
• “The fused value is the truth.” It is an estimate with a confidence. Keep both, plus the lineage.

§05How Kimo uses data fusion

In Kimo Defense Intelligence, the tracks model fuses ADS-B, OpenSky and AIS observations with provenance on every point, so the Airspace map shows both a position and the reason to trust it. The same principle drives Kimo’s business products: the semantic layer resolves customers across CRM, billing and product data before computing a single metric.

§06Related terms

• Multilateration (MLAT): an independent source that makes fusion worthwhile.
• OSINT: fusing open sources into an assessed picture.
• Data model: the entities that fused observations attach to.