Key points are not available for this paper at this time.
We present a novel computational camera with high spatial, spectral, and temporal resolution as a practical tool for real-world applications in VIS-NIR hyperspectral (HSI) video imaging. Hyperspectral images contain orders of magnitude more information than colour (RGB) images and promise to reveal new insights in many application domains. However, widespread adoption of HSI has been hindered by the performance limitations of imaging equipment, complexity of application development and costs of deployment. We report on a compressed sensing approach using a variation of coded aperture snapshot spectral imaging (CASSI) with low aberration dispersive optics to capture the multiplexed projection of spatial-spectral scene information on standard CMOS technology. We overcome the typically slow computations associated with CASSI signal reconstruction and report state-of-the-art performance using machine learning methods for signal unmixing at speeds enabling live processing on networked NVIDIA gpu-enabled platforms. Contemporary applications and the motivation for packaging as a computer vision development kit are discussed.
Chappell et al. (Mon,) studied this question.