Sub-pixel image translation estimation on a nanosatellite platform

[摘要] ENGLISH ABSTRACT: Nanosatellites are limited in their physical size, which limits the physical sizeof payloads they can carry, thereby limiting the quality of images taken duringCubeSat Earth Observation missions. Algorithms exist that combine partiallyoverlapping images to produce better output image quality. These algorithmsmay either improve the signal-to-noise ratio via averaging, increase resolutionvia super-resolution or merely remove redundant information via mosaicing.Typically, they only function properly if the geometric transformations betweenconsecutive images are known with high accuracy. They can either be appliedterrestrially or on-board a satellite. Downloading large raw image data sets forterrestrial processing is impractical for a CubeSat mission, and therefore anon-board solution is desirable.This thesis discusses the accurate determination of the transformationbetween consecutive images on-board, laying the foundation for e cient onboardde-noising, super-resolution and mosaicing. Two common methodsused to determine translation { normalised cross correlation (NCC) and phasecorrelation { are investigated. From simulated results, NCC is shown to bethe better candidate for our application. NCC achieves sub-pixel accuracyby making use of polynomial least squares regression. NCC is well suited forimplementation on a satellite platform where images are captured in quicksuccession, resulting in partially overlapping images with little rotation betweenframes.We compare two potential hardware platforms { the MicroZed 7020 andJetson TK1 { and then describe how we implemented our proposed solutiononto the former, using a hardware description language. Software simulationandrmware-implementation results, using simulated data, are comparedand discussed. Subsequently, the MicroZed 7020's implemented design ischaracterised, compared and discussed in terms of algorithm and platformperformance.

[发布机构] Stellenbosch University