Desenvolvimento de um método semiautomático para geração de ground truths de vídeos

• Autor(a) principal: Gaspari, Tiago De [UNESP]
• Data de Publicação: 2015
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/138444; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/08-04-2016/000863591.pdf
• topic: Computação; Processamento de imagens - Tecnicas digitais; Vídeo digital; Algoritmos de computador; Visão por computador; Computer science

Several bilayer video segmentation algorithms, for the extraction of elements of interest in the foreground (usually people) in uncontrolled environments, have been proposed for various applications such as Augmented Reality systems, video chats or for video compression. To analyze the quality of the videos generated by the segmentation algorithms, different methods compare them with their ground truths, which consist of references of the best possible segmentation of a video. Usually this ground truth is obtained manually, so the user may have to segment each frame of the video. Of course this process is laborious, time consuming and frequently not performed for all frames of the video. Because of this, the ground truth, usually, has low resolution and short duration. These aspects can be an issue to the effectiveness of using these ground truths in the segmentation quality evaluation process. In this context, this project had as its main focus the development of a semi-automatic method for the generation of ground truths of videos, using depth information, in order to minimize the user interaction, the implementation process of this method as a tool that combines automatic and interactive steps, and the results, that are compared to other studies in literature