Software design for dynamic stitching of multi-spectral images of field crops

We proposed an alternate configuration for dynamic multi-spectral imaging of plants at ground level, using a one-sensor monochrome FireWire-A camera combined with a liquid crystal tunable filter (LCTF) tunable at 760 nm and 695 nm. We developed an algorithm based on the traditional NDVI procedure to untraditionally solve the problem of image correspondence within sequences of spectral images collected from a moving platform. This algorithm was designed from a multi-threaded software engineering approach and had been shown to work well during acceleration, steady, and deceleration phases of a moving platform under laboratory conditions and one limited field test. The algorithm had two components: one to determine the optimal number of pixel rows passing by the camera viewport during the time period needed for the LCTF to switch between wavelengths; the other to determine the proper conditions to trigger an image saving event so as to minimize the number of images recorded that would however have enough overlaps to permit their assembly into potential field-level maps. The proposed system could process up to 14 image sets per second (3 images per set-760 nm, 695 nm and NDVI) and was found to tolerate light breezes under field conditions.