Robust Dynamic Multi-Modal Data Fusion: A Model Uncertainty Perspective

TL;DR

This paper introduces a robust multi-modal data fusion framework that effectively handles modality failures in nonlinear, non-Gaussian dynamic systems using a model uncertainty perspective and a dynamic model averaging particle filter.

Contribution

It proposes a novel model uncertainty approach with a DMA-based particle filter that manages all modality combinations efficiently, improving robustness over existing methods.

Findings

Outperforms state-of-the-art methods in experiments

Handles multiple modality failures effectively

Maintains computational efficiency for small modality sets

Abstract

This paper is concerned with multi-modal data fusion (MMDF) under unexpected modality failures in nonlinear non-Gaussian dynamic processes. An efficient framework to tackle this problem is proposed. In particular, a notion termed modality "\emph{usefulness}", which takes a value of 1 or 0, is used for indicating whether the observation of this modality is useful or not. For $n$ modalities involved, $2^n$ combinations of their "\emph{usefulness}" values exist. Each combination defines one hypothetical model of the true data generative process. Then the problem of concern is formalized as a task of nonlinear non-Gaussian state filtering under model uncertainty, which is addressed by a dynamic model averaging (DMA) based particle filter (PF) algorithm. This DMA algorithm employs $2^n$ models, while all models share the same state-transition function and a unique set of particle values. That makes its computational complexity only slightly larger than a single model based PF algorithm, especially for scenarios in which $n$ is small. Experimental results show that the proposed solution outperforms remarkably state-of-the-art methods. Code and data are available at https://github.com/robinlau1981/fusion.