Variational Voxel Pseudo Image Tracking

TL;DR

This paper introduces a variational neural network approach to 3D single object tracking that estimates uncertainty, improving tracking accuracy by leveraging uncertainty-aware feature correlation.

Contribution

It presents a novel Variational VPIT method with an uncertainty-aware cross-correlation module for enhanced 3D object tracking.

Findings

Both proposed methods improve tracking performance.

Penalizing uncertain features yields the best uncertainty quality.

Uncertainty estimation enhances model reliability in critical applications.

Abstract

Uncertainty estimation is an important task for critical problems, such as robotics and autonomous driving, because it allows creating statistically better perception models and signaling the model's certainty in its predictions to the decision method or a human supervisor. In this paper, we propose a Variational Neural Network-based version of a Voxel Pseudo Image Tracking (VPIT) method for 3D Single Object Tracking. The Variational Feature Generation Network of the proposed Variational VPIT computes features for target and search regions and the corresponding uncertainties, which are later combined using an uncertainty-aware cross-correlation module in one of two ways: by computing similarity between the corresponding uncertainties and adding it to the regular cross-correlation values, or by penalizing the uncertain feature channels to increase influence of the certain features. In experiments, we show that both methods improve tracking performance, while penalization of uncertain features provides the best uncertainty quality.