UCL Discovery
UCL home » Library Services » Electronic resources » UCL Discovery

Applications of machine vision to cloud studies using stereoscopic satellite images

Shin, Dongseok; (1995) Applications of machine vision to cloud studies using stereoscopic satellite images. Doctoral thesis (Ph.D.), University College London (United Kingdom). Green open access

[img] Text
Applications_of_machine_vision.pdf

Download (18MB)

Abstract

This thesis describes three main fields of study for ATSR image processing in order to determine cloud-top height using stereoscopic ATSR images. Firstly, several algorithms are proposed to reduce the geometric errors of navigation of ATSR images to a sub-pixel level. A navigation model is derived to correct elliptically-distorted raw ATSR images to a pixel level geo-location accuracy. The non-systematic geometric errors are analysed and corrected further by a fine co-location algorithm and a precision correction algorithm. The fine co-location algorithm detects and matches coastline sequences in a pair of stereoscopic ATSR images. It employs binary cross-correlation techniques and dynamic programming for the automated coastline matching. A polynomial warping technique reduces the co-location error between the nadir and forward images to a sub-pixel level. The precision correction algorithm refines the navigation model by estimating the unknown geometric error sources using a Kalman filter. This algorithm improves the geo-location accuracy and provides accurate height reference for the determination of the cloud-top height using a stereo vision technique. A new algorithm, using a region segmentation technique, is proposed to achieve fast and accurate cloud detection from ATSR infra-red images over ocean. The algorithm includes a spatial variability test using a segmentation technique, which gives faster and more accurate results than the conventional spatial coherence test. An accurate temperature threshold between clouds and clear sea can be determined directly from the segmented image and hence it detects many warm and/or partially filled cloud pixels which cannot be detected by the conventional IR gross test with pre-fixed thresholds. Finally, disparities are determined from the stereoscopic ATSR cloud images by using automatic stereo matching algorithms. The performance of different matching algorithms is compared. An object-based matching algorithm is suggested for the ATSR IR cloud image matching. The navigation model derives the height of clouds from the disparities which are determined by stereo matching algorithms. All developed algorithms result in fast and fully automatic operation for ATSR image processing to determine cloud-top heights.

Type: Thesis (Doctoral)
Qualification: Ph.D.
Title: Applications of machine vision to cloud studies using stereoscopic satellite images
Open access status: An open access version is available from UCL Discovery
Language: English
Additional information: Thesis digitised by ProQuest.
Keywords: (UMI)AAI10045732; Applied sciences; Cloud; Satellite images
URI: https://discovery.ucl.ac.uk/id/eprint/10099806
Downloads since deposit
24Downloads
Download activity - last month
Download activity - last 12 months
Downloads by country - last 12 months

Archive Staff Only

View Item View Item