Enhanced visual object tracking in division-of-focal-plane imagery using a joint polarization descriptor
Applied Optics, vol.65, no.16, pp.5315-5329, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 65 Issue: 16
- Publication Date: 2026
- Doi Number: 10.1364/ao.592531
- Journal Name: Applied Optics
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, EMBASE, MEDLINE
- Page Numbers: pp.5315-5329
- Hacettepe University Affiliated: Yes
Abstract
Polarization imaging provides physical scene cues beyond conventional intensity images, enabling improved characterization of surface geometry and material properties. The degree of linear polarization (DoLP) and the angle of linear polarization (AoLP) encode complementary information related to polarization magnitude and orientation, which are linked to surface normal and reflectance behavior. However, existing polarization-aware descriptors for division-of-focal-plane (DoFP) cameras mainly rely on gradient-based formulations and emphasize DoLP, while the coupling between AoLP and DoLP remains largely unexplored. This paper introduces what we believe to be a novel polarization feature descriptor, the joint histogram of AoLP and DoLP (JHAD), for real-time object tracking in DoFP imagery. JHAD explicitly models the joint distribution of polarization magnitude and orientation within a unified framework, preserving discriminative polarization characteristics lost in separate representations. The proposed descriptor is integrated into multiple correlation filter-based trackers and evaluated against conventional features and state-of-the-art polarization descriptors. Additionally, its integration into a Siamese tracking framework is investigated. To address the lack of suitable benchmarks, a new DoFP polarization tracking dataset consisting of 105 sequences with challenging conditions, including heavy occlusion, is introduced. Experimental results demonstrate that JHAD consistently improves tracking performance in terms of success and precision across all evaluated frameworks while maintaining real-time capability. These findings highlight the effectiveness of jointly modeling AoLP and DoLP for robust polarization-based object tracking in complex scenes.