Multi-target particle filtering for the probability hypothesis density

TL;DR

This paper introduces a particle filter implementation of the probability hypothesis density (PHD) filter for multi-target tracking, effectively handling the computational challenges and data association issues in non-linear, multi-vehicle terrain tracking scenarios.

Contribution

It presents a novel particle filter approach for the PHD filter, enabling efficient and robust multi-target tracking without data association.

Findings

Successfully tracks a changing number of vehicles

Achieves near-real-time performance

Demonstrates robustness in non-linear terrain tracking

Abstract

When tracking a large number of targets, it is often computationally expensive to represent the full joint distribution over target states. In cases where the targets move independently, each target can instead be tracked with a separate filter. However, this leads to a model-data association problem. Another approach to solve the problem with computational complexity is to track only the first moment of the joint distribution, the probability hypothesis density (PHD). The integral of this distribution over any area S is the expected number of targets within S. Since no record of object identity is kept, the model-data association problem is avoided. The contribution of this paper is a particle filter implementation of the PHD filter mentioned above. This PHD particle filter is applied to tracking of multiple vehicles in terrain, a non-linear tracking problem. Experiments show that the filter can track a changing number of vehicles robustly, achieving near-real-time performance.