Polarimetric Pose Prediction

TL;DR

This paper demonstrates that incorporating polarisation information into pose estimation models significantly enhances accuracy, especially for reflective and transparent objects, by combining physical priors with data-driven learning.

Contribution

It introduces a hybrid model leveraging polarisation data and physical priors for improved 6D pose prediction, and provides a new multi-modal dataset for benchmarking.

Findings

Improved pose accuracy over state-of-the-art methods.

Effective estimation for reflective and transparent objects.

Introduction of a new annotated multi-modal pose dataset.

Abstract

Light has many properties that vision sensors can passively measure. Colour-band separated wavelength and intensity are arguably the most commonly used for monocular 6D object pose estimation. This paper explores how complementary polarisation information, i.e. the orientation of light wave oscillations, influences the accuracy of pose predictions. A hybrid model that leverages physical priors jointly with a data-driven learning strategy is designed and carefully tested on objects with different levels of photometric complexity. Our design significantly improves the pose accuracy compared to state-of-the-art photometric approaches and enables object pose estimation for highly reflective and transparent objects. A new multi-modal instance-level 6D object pose dataset with highly accurate pose annotations for multiple objects with varying photometric complexity is introduced as a benchmark.