Calibrated Vehicle Paint Signatures for Simulating Hyperspectral Imagery

• Main Authors: Mulhollan, Zachary, Rangnekar, Aneesh, Bauch, Timothy, Hoffman, Matthew J, Vodacek, Anthony
• Format: Journal Article
• Language: English
• Published: 16-04-2020
• DOI: 10.48550/arxiv.2004.08228

We investigate a procedure for rapidly adding calibrated vehicle visible-near
infrared (VNIR) paint signatures to an existing hyperspectral simulator - The
Digital Imaging and Remote Sensing Image Generation (DIRSIG) model - to create
more diversity in simulated urban scenes. The DIRSIG model can produce
synthetic hyperspectral imagery with user-specified geometry, atmospheric
conditions, and ground target spectra. To render an object pixel's spectral
signature, DIRSIG uses a large database of reflectance curves for the
corresponding object material and a bidirectional reflectance model to
introduce s due to orientation and surface structure. However, this database
contains only a few spectral curves for vehicle paints and generates new paint
signatures by combining these curves internally. In this paper we demonstrate a
method to rapidly generate multiple paint spectra, flying a drone carrying a
pushbroom hyperspectral camera to image a university parking lot. We then
process the images to convert them from the digital count space to spectral
reflectance without the need of calibration panels in the scene, and port the
paint signatures into DIRSIG for successful integration into the newly rendered
sets of synthetic VNIR hyperspectral scenes.