Aceleração de algoritmos de sumarização de vídeos com processadroes gráficos (GPUs) e multicore CPUs
The recent and fast evolution of digital media have stimulated the creation, storage and distribution of data, such as digital videos, generating a large volume of data and requiring efficient technologies to increase the usability of these data. Video summarization methods consist of generating con...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis de maestría |
| Estado: | Versión publicada |
| Fecha de publicación: | 2014 |
| País: | Brasil |
| Institución: | Universidade Federal de Minas Gerais (UFMG) |
| Repositorio: | Repositório Institucional da UFMG |
| Idioma: | portugués |
| OAI Identifier: | oai:repositorio.ufmg.br:1843/ESBF-9TENPA |
| Acceso en línea: | http://hdl.handle.net/1843/ESBF-9TENPA |
| Access Level: | acceso abierto |
| Palabra clave: | Processamento de vídeos Sumarização de vídeos GPUs Multicore CPUs Algoritmos paralelos Computação Processamento de imagens Técnicas digitais |
| Sumario: | The recent and fast evolution of digital media have stimulated the creation, storage and distribution of data, such as digital videos, generating a large volume of data and requiring efficient technologies to increase the usability of these data. Video summarization methods consist of generating concise summaries of video contents and it enable faster browsing, indexing and accessing of large video collections. However, these methods often perform slow with large duration and high quality video data. One way to reduce this long time of execution is to develop parallel algorithms, using the advantages of the recent computer architectures that allow high parallelism, i.e., Graphics Processor Units (GPUs) and multicore CPUs. This work proposes parallelizations of two video summarization methods. The former is based on color feature extraction from video frames and k-means clustering algorithm and the latter is based on temporal video segmentation and visual words obtained by local descriptors. For the two methods, some implementations were considered: GPUs, multicore CPUs, and ultimately a distribution of computations steps onto both hardware to maximise performance. The experiments were performed using 240 videos varying frame resolution (320 X 240, 640 X 360, 1280 X 720 e 1920 X1080 pixels) and video length (1,3,5,10,20 and 30 minutes). The results shows that the implementations overcome the sequential version of both methods, keeping the quality of the summaries. |
|---|