In English

Realization of efficient change detection in SAR images

Joanna Eriksson
Göteborg : Chalmers tekniska högskola, 2015. 71 s.
[Examensarbete på avancerad nivå]

There exist algorithms for detecting if a change has taken place in a land area by analyzing images over the land from two occasions. The process of detecting changes using these algorithms is called change detection. The current MATLAB implementation of the algorithms are too slow for running in a real-time environment. This thesis has the purpose of determining whether a real-time performance would be possible by translating from MATLAB to C++ and parallelizing. The parallelization is performed with the API OpenMP and FFT (Fast Fourier Transform) computations are done with the C library FFTW. It was not previously known if it is possible to run this change detection implementation in real-time. Similar research has studied this problem for other types of images and other types of algorithms for change detection. The considered algorithms of change detection can be divided into five parts and within the time frame of the project there has been time to thorughly study the first step, called pie-filter or spatial band-pass filter and to do a brief and theoretical study of the second step, matching. The speedup of the pie-filter is shown to be about 9 times compared to the original MATLAB code. The maximum possible total execution time is the time it takes to take a SAR image of the ground. The pie-filter is executing in 10% of the maximum possible total execution time of the real-time change detection. This is a result that indicates a possible real-time execution. The report also contains an evaluation of different methods for performing FFT using FFTW. These methods are analyzed and it is found that for different levels of parallelization and input size, different methods are optimal to use. The results regarding FFT are important because they can be applied to other applications.



Publikationen registrerades 2015-06-29. Den ändrades senast 2015-06-29

CPL ID: 219084

Detta är en tjänst från Chalmers bibliotek