Authors:

Priyank Behera

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Kalman-filter-based multi-target tracking is fundamental for many modern sensor-fusion and military architectures. However, conventional Kalman updates are notoriously fragile and inaccurate in the presence of colored noise. These errors can compound on each other and ultimately lead to filter divergence. This paper proposes a numerically robust alternative using Gaussian Influence Diagrams (GIDs). By reframing the Kalman filter as a sequence of local Bayesian decision problems, the GID framework preserves conditional independencies and enables update equations expressed solely through factorizations. This research benchmarks the proposed Influence Diagram JPDAF (ID-JPDAF) against the classical Joint Probabilistic Data Association Filter (JPDAF). We generated several simulated multi-target radar scenarios using a nearly constant-velocity model. Performance was assessed over 1000 Monte- Carlo trials using root-mean-square error, track continuity, and covariance-positivity metrics. For track association and detection metrics, we used GOSPA localization score and SIAP. In addition, we tested a data set featuring colored measurement noise to validate real-world applicability. ID-JPDAF was more accurate with this dataset, and its performance across standard multi-target tracking metrics were not worse than a classical JPDAF. The findings suggest that influence- diagram reasoning offers a path to numerically robust multi-target tracking for next-generation defense systems. Keywords: Data Association; Bayesian Statistics; Kalman Filtering; Multi-Target Tracking

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Priyank Behera