Detecção de Mudança

Informações sobre alterações em uma cena desempenham um papel central em uma infinidade de aplicações. Gestão de desastres, crescimento urbano, segurança, áreas queimadas e vigilância são exemplos de problemas que o conhecimento das mudanças locais é essencial. Em geral, o monitoramento de vídeo realizado por seres humanos é propenso a erros, devido à conhecida falta de atenção em tarefas repetitivas que podem levar muitas horas. A monitoração manual de alterações é muito trabalhosa e muitas vezes impraticável, resultando em amostragem de dados inconsistente e pouco frequente, bem como erro humano. A tarefa básica de detecção de alterações nas imagens é encontrar pixels (ou regiões) de uma imagem de referência que sejam diferentes das outras imagens. Mudanças significativas podem incluir a aparência ou o desaparecimento de objetos, o movimento de objetos e mudanças de forma nos objetos. Por outro lado, as variações na imagem causadas pelo movimento da câmera, ruído, mudanças de iluminação, atenuação não uniforme, absorção atmosférica, árvores balançando, água ondulada ou monitores piscando, devem ser ignoradas pelo método de detecção de mudança.

Publicações Relacionadas

Marco Tulio Alves Rodrigues

Detecção de Mudanças em Cenas Terrestres usando Imagens Aéreas Tese PhD

2016.

Resumo | Links | BibTeX

@phdthesis{Rodrigues:2016:PhD,
title = {Detecção de Mudanças em Cenas Terrestres usando Imagens Aéreas},
author = {Marco Tulio Alves Rodrigues},
url = {http://smartsenselab.dcc.ufmg.br/wp-content/uploads/2019/03/dissertation_2016_MarcoTulio.pdf},
year = {2016},
date = {2016-03-22},
abstract = {This study addresses the problem of change detection in landscapes using aerial images acquired at different times, important for many applications. The monitoring pipeline, for instance, the usual way to change detection is a task performed by human operators which evaluates a video of a monitoring camera and searches for changes in the scene from the comparison of image pairs. This procedure is prone to errors because it is a tedious task, therefore is a justification to automated method. Besides enabling the reduction of errors and speed up the monitoring process, a system automatic can be used as a filter to provide a set of key frames that should receive more attention from the operator. Thus, the system can help the operators on the decision making process regarding the actions to be performed. The basic procedure for detecting changes is to find a set of pixels or regions that are different in another test image. However, images acquired at different dates may be influenced by radiometric and registration factors. In other words, the influence of camera movement, lighting variation, and atmospheric variation must be minimized. One of the methods proposed extracts local descriptors in the image blocks and provides an estimate of change using a non-parametric model (KDE). Unlike background subtraction and remote sensing methods which are based on pixels and assume independence between them, the proposed approach not requires a complex learning phase and it is capable of detecting changes using only two images. The second method applies an image segmentation before make the matching of the similar regions. In the experiments, the proposed approaches are compared to techniques used in change detection. According to the results, the proposed approach based on non-parametric model outperforms other methods found in the literature, mainly due to the fact that the approach is more robust to lighting variation. The results also demonstrate that the approach is able to filter images that should be further analyzed by operators.},
keywords = {},
pubstate = {published},
tppubtype = {phdthesis}
}

Fechar

This study addresses the problem of change detection in landscapes using aerial images acquired at different times, important for many applications. The monitoring pipeline, for instance, the usual way to change detection is a task performed by human operators which evaluates a video of a monitoring camera and searches for changes in the scene from the comparison of image pairs. This procedure is prone to errors because it is a tedious task, therefore is a justification to automated method. Besides enabling the reduction of errors and speed up the monitoring process, a system automatic can be used as a filter to provide a set of key frames that should receive more attention from the operator. Thus, the system can help the operators on the decision making process regarding the actions to be performed. The basic procedure for detecting changes is to find a set of pixels or regions that are different in another test image. However, images acquired at different dates may be influenced by radiometric and registration factors. In other words, the influence of camera movement, lighting variation, and atmospheric variation must be minimized. One of the methods proposed extracts local descriptors in the image blocks and provides an estimate of change using a non-parametric model (KDE). Unlike background subtraction and remote sensing methods which are based on pixels and assume independence between them, the proposed approach not requires a complex learning phase and it is capable of detecting changes using only two images. The second method applies an image segmentation before make the matching of the similar regions. In the experiments, the proposed approaches are compared to techniques used in change detection. According to the results, the proposed approach based on non-parametric model outperforms other methods found in the literature, mainly due to the fact that the approach is more robust to lighting variation. The results also demonstrate that the approach is able to filter images that should be further analyzed by operators.

Fechar

Marco Tulio Alves Rodrigues; Daniel Balbino de Mesquita; Erickson R Nascimento; William Robson Schwartz

Change detection based on feature invariant to monotonic transforms and spatially constrained matching Journal Article

Em: Journal of Electronic Imaging, 25 (1), pp. 1-10, 2016.

Links | BibTeX

Marco Tulio Alves Rodrigues; Daniel Balbino; Erickson R Nascimento; William Robson Schwartz

A Non-Parametric Approach to Detect Changes in Aerial Images Inproceedings

Em: 14th Iberoamerican Congress on Pattern Recognition (CIARP), pp. 1-8, 2015.

Links | BibTeX

Marco Tulio Alves Rodrigues; L O Milen; E R Nascimento; William Robson Schwartz

Change detection based on features invariant to monotonic transforms and spatial constrained matching Inproceedings

Em: Acoustics, Speech and Signal Processing (ICASSP), 2014 IEEE International Conference on, pp. 4334-4338, 2014.

Links | BibTeX