Conditional Measurement Density Estimation in Sequential Monte Carlo via Normalizing Flow

TL;DR

This paper introduces a method to learn valid and expressive measurement models in sequential Monte Carlo methods using conditional normalizing flows, improving accuracy and training speed in visual tracking tasks.

Contribution

It presents a novel approach to construct valid probability densities for measurement models via conditional normalizing flows within differentiable particle filters.

Findings

Improved estimation accuracy in visual tracking.

Faster training convergence compared to previous methods.

Valid probability densities enhance measurement uncertainty quantification.

Abstract

Tuning of measurement models is challenging in real-world applications of sequential Monte Carlo methods. Recent advances in differentiable particle filters have led to various efforts to learn measurement models through neural networks. But existing approaches in the differentiable particle filter framework do not admit valid probability densities in constructing measurement models, leading to incorrect quantification of the measurement uncertainty given state information. We propose to learn expressive and valid probability densities in measurement models through conditional normalizing flows, to capture the complex likelihood of measurements given states. We show that the proposed approach leads to improved estimation performance and faster training convergence in a visual tracking experiment.