|
|
Exploitation of GPU in graphics and image processing algorithms
Jošth, Radovan ; Svoboda, David (oponent) ; Trajtel,, Ľudovít (oponent) ; Herout, Adam (vedoucí práce)
This thesis introduces several selected algorithms, which were primarily developed for CPUs, but based on high demand for improvements; we have decided to utilize it on behalf of GPGPU. This modification was at the same time goal of our research. The research itself was performed on CUDA enabled devices. The thesis is divided in accordance with three algorithm’s groups that have been researched: a real-time object detection, spectral image analysis and real-time line detection. The research on real-time object detection was performed by using LRD and LRP features. Research on spectral image analysis was performed by using PCA and NTF algorithms and for the needs of real-time line detection, we have modified accumulation scheme for the Hough transform in two different ways. Prior to explaining particular algorithms and performed research, GPU architecture together with GPGPU overview are provided in second chapter, right after an introduction. Chapter dedicated to research achievements focus on methodology used for the different algorithm modifications and authors’ assess to the research, as well as several products that have been developed during the research. The final part of the thesis concludes our research and provides more information about the research impact.
|
|
|
Exploitation of GPU in graphics and image processing algorithms
Jošth, Radovan ; Svoboda, David (oponent) ; Trajtel,, Ľudovít (oponent) ; Herout, Adam (vedoucí práce)
This thesis introduces several selected algorithms, which were primarily developed for CPUs, but based on high demand for improvements; we have decided to utilize it on behalf of GPGPU. This modification was at the same time goal of our research. The research itself was performed on CUDA enabled devices. The thesis is divided in accordance with three algorithm’s groups that have been researched: a real-time object detection, spectral image analysis and real-time line detection. The research on real-time object detection was performed by using LRD and LRP features. Research on spectral image analysis was performed by using PCA and NTF algorithms and for the needs of real-time line detection, we have modified accumulation scheme for the Hough transform in two different ways. Prior to explaining particular algorithms and performed research, GPU architecture together with GPGPU overview are provided in second chapter, right after an introduction. Chapter dedicated to research achievements focus on methodology used for the different algorithm modifications and authors’ assess to the research, as well as several products that have been developed during the research. The final part of the thesis concludes our research and provides more information about the research impact.
|
|
|
Implementace obrazových klasifikátorů v FPGA
Kadlček, Filip ; Puš, Viktor (oponent) ; Fučík, Otto (vedoucí práce)
Práce je zaměřena na obrazové klasifikátory a jejich implementaci v FPGA. Klasifikátory dělí na dvě skupiny - slabé a silné klasifikátory. Ve skupině silných klasifikátorů se zaměřuje především na AdaBoost. Ve skupině slabých klasifikátorů jsou probrány základní příznakové klasifikátory, jakými jsou například klasifikátory založené na Haarových nebo Gaborových vlnkách, ale především je kladen důraz na klasifikátory LBP, LRP a LR. Naposled uvedené klasifikátory jsou vhodné pro implementaci v FGPA. Na základě těchto klasifikátorů je navržena pseudo-paralelní architektura. Architektura uvažuje provedení klasifikace v FPGA a následné zpracovávání výsledků v počítači. Navržený klasifikátor je velmi rychlý a každý hodinový cyklus produkuje výstup klasifikace.
|
host ::
RSS.