Hyperspectral scene analysis via structure from motion

Abstract

We present an overview of a structure from motion (SFM) pipeline for processing hyperspectral imagery (HSI), and demonstrate the data exploitation advantages associated with post-processing HSI data in a 3D environment. Using only raw HSI datacubes as input, we leverage HSI anomaly detection and spectral matching to create a 3D spatial model of the scene being imaged. The resulting 3D space provides an intuitive basis for all forms of HSI analysis. We demonstrate the usefulness of the proposed HSI SFM pipeline through an experimental data set collected using an aerial hyperspectral sensor.