dataassociation

Data association, sometimes encountered as dataassociation, is the process of linking observations or measurements to the correct underlying sources or targets, and more broadly of fusing and reconciling data from multiple sensors or records. It is a core problem in fields such as multi-target tracking, sensor fusion, data integration, and entity resolution. The goal is to determine which measurements originate from the same object or entity and which belong to different ones, while accounting for noise, clutter, missed detections, and measurement uncertainty.

In tracking and fusion, data association decisions are essential to maintain consistent object identities over time. In entity resolution, data association corresponds to linking records that refer to the same real-world entity across databases.

Common approaches include probabilistic data association methods such as PDA and JPDA, which compute probabilities that each measurement is associated with each target, and multi-hypothesis tracking (MHT), which maintains several association hypotheses and updates them as new data arrives. Optimization-based methods include the Hungarian algorithm for one-to-one data association, and other combinatorial solvers like auction algorithms. Gating and likelihood ratios are often used to prune implausible associations. Track management and context information may be incorporated to improve robustness.

Applications of data association span radar, air traffic control, autonomous and assisted driving, video surveillance, and data integration tasks such as record linkage and entity resolution. Key challenges include high data volumes, real-time processing requirements, clutter and missed detections, highly similar targets or records, and sensor or database heterogeneity.