Auto-Calibrated 3D Hyperspectral Scanning Using a Heterogeneous Set of Cameras and Lights with Spectrally-Optimal Next-Best-View Planning

Abstract

Hyperspectral imaging generally requires specialized hardware and well-calibrated experimental setups. As a result, combining hyperspectral imaging with 3D scanning is an arduous task that has not yet seen widespread adoption by the automation and robotics communities. In this paper, we present a streamlined method of calibrating and operating a cost-effective multi-camera 3D hyperspectral scanner that is suitable for field applications. The design uses off-the-shelf (uncalibrated) cameras. We analyze spectral estimates of common reflectance curves and comparing our fast and locally-refined next best view sequences to globally optimal sequences. The characterization of lighting conditions and camera sensitivity curves are demonstrated through alternating least squares. The results are compared to the ground truth sensitivity and spectral power distribution curves.