Regional variance for multi-object filtering

Emmanuel Delande, Murat Üney, Jérémie Houssineau, Daniel E. Clark

Abstract

Recent progress in multi-object filtering has led to algorithms that compute the first-order moment of multi-object distributions based on sensor measurements. The number of targets in arbitrarily selected regions can be estimated using the first-order moment. In this work, we introduce explicit formulae for the computation of the second-order statistic on the target number. The proposed concept of regional variance quantifies the level of confidence on target number estimates in arbitrary regions and facilitates information-based decisions. We provide algorithms for its computation for the Probability Hypothesis Density (PHD) and the Cardinalized Probability Hypothesis Density (CPHD) filters. We demonstrate the behaviour of the regional statistics through simulation examples.

Introduction

In multi-target tracking applications such as radar surveillance or autonomous navigation, the number of objects in a scene is typically unknown and dynamic. Standard filtering techniques like the Probability Hypothesis Density (PHD) and Cardinalized PHD (CPHD) filters have gained popularity due to their ability to efficiently estimate the "first-order moment"—the expected number of targets in a given region.

However, for safety-critical tasks like automated collision avoidance or sensor management, knowing the mean estimate is insufficient. Engineers require a measure of the estimate's reliability. While previous research addressed the variance of the target count for the entire surveillance area, explicit formulas for localized, regional variance were missing. This paper fills that gap by providing a measure of local uncertainty, allowing trackers to quantify confidence in target counts within any arbitrary sub-region.

Methodology

The authors employ a measure-theoretical framework based on Random Finite Sets (RFS) to derive these second-order statistics. The derivation utilizes Probability Generating Functionals (PGFl) and Laplace functionals to characterize the point process distribution. By applying functional differentiation—specifically "chain differentials"—the authors extract explicit update equations for regional variance.

As illustrated in Figure 2, the regional variance is integrated into the standard filtering cycles. A key insight of this work is that while filters often assume simple distributions (Poisson or i.i.d.) for prediction, the updated process after measurement is more complex. Calculating variance immediately after the update preserves vital uncertainty information before the distribution is "collapsed" back into a simpler form for the next iteration.

Conceptual flow of CPHD and PHD filters incorporating variance updates through I.I.D. and Poisson approximations.

Experimental Setup

The proposed methods were evaluated using a simulated 2D multi-target scenario. A range-bearing sensor at the origin tracks five targets moving with linear constant velocity. The scenario features crossing trajectories to specifically test how the filters handle proximity-induced uncertainty. Clutter is modeled as a Poisson process with a rate of 20 points per scan. The simulations specifically examine the impact of varying detection probabilities ($P_D = 0.85, 0.90, 0.95$) on regional uncertainty.

Key Results and Discussion

The study successfully derived closed-form expressions for regional statistics in both PHD and CPHD filters. Simulations confirmed that regional variance provides a mathematically sound "confidence interval" around target estimates. As shown in Figure 6, regional variance is highly sensitive to sensor quality; as the probability of detection ($P_D$) decreases, the variance increases, reflecting lower confidence in the tracker's estimates.

Regional variance over time for different detection probabilities ($P_D$), showing spikes during target crossings (a) and general uncertainty levels (b).

Furthermore, regional variance serves as a proxy for target resolution. In the crossing trajectories scenario, the variance spikes at the moments of closest proximity (e.g., around 110s in Figure 6a). This indicates that the filter is uncertain whether it is observing a single merged target or multiple distinct objects.

Regarding computational efficiency, the PHD filter's variance update remains linear $O(m)$ with respect to the number of measurements $m$. The CPHD filter, however, exhibits a higher complexity of $O(m^3 \log^2 m)$ because it must process pairs of measurements to compute second-order statistics.

Conclusion

By introducing regional variance, this work provides a rigorous method for quantifying local uncertainty in multi-object tracking. This capability allows for more informed decision-making in complex environments. Future research will focus on reducing the computational overhead for the CPHD filter and exploring higher-order statistics, such as skewness and kurtosis, to further refine error modeling in multi-target systems.

Summary updated

Abstract

Recent progress in multi-object filtering has led to algorithms that compute the first-order moment of multi-object distributions based on sensor measurements. The number of targets in arbitrarily selected regions can be estimated using the first-order moment. In this work, we introduce explicit formulae for the computation of the second-order statistic on the target number. The proposed concept of regional variance quantifies the level of confidence on target number estimates in arbitrary regions and facilitates information-based decisions. We provide algorithms for its computation for the Probability Hypothesis Density (PHD) and the Cardinalized Probability Hypothesis Density (CPHD) filters. We demonstrate the behaviour of the regional statistics through simulation examples.
BibTeX
@article{delande2014regional,
  author = {Emmanuel Delande and Murat Üney and Jérémie Houssineau and Daniel E. Clark},
  title = {{Regional variance for multi-object filtering}},
  journal = {IEEE Trans. Signal Process.},
  year = {2014},
  url = {https://openreview.net/forum?id=y4jsXC4gGh},
}
Figure