Distributed multi-view multi-target tracking based on CPHD filtering

TL;DR

This paper introduces a novel distributed multi-target tracking method that effectively fuses sensor data with unreliable and time-varying fields-of-view using CPHD filtering, clustering, and multi-Bernoulli approximations.

Contribution

It proposes a new fusion algorithm that handles unknown, unreliable, and dynamic sensor fields-of-view by decomposing intensity functions and applying GCI or AA fusion rules.

Findings

The method improves tracking accuracy in challenging FoV conditions.

Simulation results demonstrate robustness against obstacles and target misdetections.

The approach outperforms existing fusion techniques in complex scenarios.

Abstract

This paper addresses distributed multi-target tracking (DMTT) over a network of sensors having different fields-of-view (FoVs). Specifically, a cardinality probability hypothesis density (CPHD) filter is run at each sensor node. Due to the fact that each sensor node has a limited FoV, the commonly adopted fusion methods become unreliable. In fact, the monitored area of multiple sensor nodes consists of several parts that are either exclusive of a single node, i.e. exclusive FoVs (eFoVs) or common to multiple (at least two) nodes, i.e. common FoVs (cFoVs). In this setting, the crucial issue is how to account for this different information sets in the fusion rule. The problem is particularly challenging when the knowledge of the FoVs is unreliable, for example because of the presence of obstacles and target misdetection, or when the FoVs are time-varying. Considering these issues, we propose an effective fusion algorithm for the case of unknown FoVs, where: i) the intensity function is decomposed into multiple sub-intensities/groups by means of a clustering algorithm; ii) the corresponding cardinality distribution is reconstructed by approximating the target random finite set (RFS) as multi-Bernoulli; and iii) fusion is performed in parallel according to either generalized covariance intersection (GCI) or arithmetic average (AA) rule. Simulation experiments are provided to demonstrate the effectiveness of the proposed approach.