Development of a Multi-spectral Camera for Computer vision applications

Abstract

The need to multispectral cameras is growing in different fields. The goal is to provide inexpensive, flexible, or high resolution acquisition set-ups for different applications. Production of cheap and easy-to-use multispectral cameras requires the design and development of specific multispectral camera sensors equipped with multispectral filter arrays. Filter arrays, band width, and spectral response of filters should be analyzed and determined before mounting the filters on the image sensor. In this study, a multispectral camera sensor was designed to be used for computer vision applications (e.g. crop/weed detection, fruit ripeness estimation etc.) covering visible range. The developed camera sensor is consisted of the image sensor, spectral filter array, a sensor board, and a driving board. A hybrid system is proposed that works using eight bands in visible range (i.e. 400-700 nm). A program was developed based on Genetic Algorithm to find the best combination of filters. The program selects the Gaussian filters using a genetic algorithm powered by wiener filter estimation method. For the selection of bands, minimum RMS of 0.0016 was obtained for the selected bands in visible. The developed camera provides eight high resolution spectral images.