Mosaic gradient histogram for object tracking in DoFP infrared polarization imaging

Abstract

This paper presents Mosaic Gradient Histogram (MGH), a feature descriptor for real-time object tracking in the video captured using Division of Focal Plane (DoFP) infrared polarization imagers. DoFP images should be demosaiced to interpolate missing polarization parameters. However, demosaicing of DoFP video is often time-consuming and subject to errors that can be amplified in the interpolation of polarization parameters. MGH encodes the spatial and polarization information into Spatial-Polarization Gradient Map (SPGM) for real-time object tracking without the need of demosaicing. Experiment results demonstrate that the proposed MGH feature descriptor outperforms conventional feature descriptors such as HOG, SIFT, HSOG, SSMHOG, LBP, and Grayscale for object tracking in DoFP infrared polarization imaging in terms of performance metrics like center accuracy and overlap precision.