A Belief Propagation-Based Track-Before-Detect Method for Objects Covering Multiple Data Cells

Abstract

In many multiobject tracking systems, including radar- and sonar-based ones, the raw sensor data are structured in data cells, with each cell representing, e.g., a different range and/or bearing value. While conventional detect-then-track algorithms convert the cell-structured data into point measurements to reduce the amount of data, track-before-detect (TBD) methods process the data directly, thus avoiding potential information loss. We propose a TBD method for the tracking of a known number of objects, each of which may cover multiple data cells. Our method relies on a new association-based data cell model that makes use of a generalized form of probabilistic data association that is able to assign multiple data cells to one object. Using this model, we develop a belief propagation-based algorithm for joint probabilistic object-cell association and object tracking. This algorithm has a moderate computational complexity in spite of the possibly high number of different object-cell association hypotheses. Numerical results demonstrate that the proposed TBD method can outperform a state-of-the-art method.