Hierarchical Association and Fusion for Multi-target Tracking with Biased Sensors

A two tier track association and fusion system is considered for the case of multiple sensors with unknown biases. In the first tier, tracks from a subset of the sensors are correlated and fused with joint bias estimation. In the second tier, fused tracks from the first tier are correlated and fused with the tracks from an additional sensor. Due to the joint fusion and bias estimation in the first tier, the multi-sensor tracks will be cross-correlated and this cross-correlation may hinder their association with the tracks held by the sensor at the second tier. The track association and fusion at the second tier will be studied and characterized for the case of unknown sensor biases at both tiers. A new definition of track density, a quantity that characterizes the difficulty of an association scene, is introduced. Since probability of detection is modeled in this analysis, each sensor may see a subset of the total number of objects in the scene. The algorithms used for track association and fusion are presented. Probability of correct association, mean squared error, and covariance consistency of the resulting tracks at the second tier are used as performance metrics.