Passive 3D Scene Reconstruction via Hyperspectral Imagery

We present the framework for a novel structure from motion (SFM) pipeline to generate 3D reconstructions of low-resolution hyperspectral imagery (HSI). Generating 3D models from a sequence of raw HSI datacubes, where each image pixel retains its spectral content of the scene, significantly expands the analysis currently possible with HSI. In addition to traditional HSI anomaly detection and spectral matching, a 3D spatial model of the scene allows for additional viewing from previously undefined viewpoints, digital elevation map generation, and enhanced object classification capabilities. State-of-the-art SFM techniques are utilized and enhanced by leveraging the spectral content recorded at each image pixel. We explore the potential of this HSI SFM pipeline using an experimental aerial data set collected using a stabilized, 160-band hyperspectral sensor on an aerial platform.