High-Resolution Hyperspectral Video Imaging Using Camera Arrays

Hyperspectral imaging systems are becoming increasingly popular because they are extending the perceivable spectral range and spectral resolution in comparison to classical color cameras which only capture a red, green, and blue channel. Hyperspectral imaging devices achieve this extension by sampling the light spectrum within a certain range for each pixel. Recording the reflected light spectrum of objects is an essential task for many detection and classification problems, since materials and processes have a unique spectral behaviour. For diverse applications, a high resolution in every dimension, that is, two spatial dimensions, the temporal dimension, and the spectral dimension, is required. Hence, the development of such a high-resolution hyperspectral video imaging system is crucial to solve challenging problems on the application side.
To tackle the challenge of high-resolution hyperspectral video imaging, four major ideas are covered in this thesis. First, the hexagonal array for hyperspectral imaging, a camera array capable of high-resolution hyperspectral video imaging, is introduced. Moreover, a mapping process of the different spectral bands is necessary in order to produce a consistent hyperspectral data cube. To achieve this, a cross spectral image registration pipeline is presented, which marks the second fundamental contribution of this work. Third, a synthetic hyperspectral video array database is introduced which enables a thorough evaluation of the cross spectral image registration pipeline. Finally, techniques on how to estimate hyperspectral images from multispectral images are introduced, which is a more cost-efficient way of providing hyperspectral images due the lower number of cameras employed.