Probabilistic object tracking for global optimization
Abstract
Object tracking in video-surveillance applications is a major topic in the signal processing and computer vision community. Its aim is to estimate properties of the imaged objects within a scene, such as their dimensions, motion, visual features, etc. Typically, object tracking is treated as an inference problem through probabilistic methods (Bayesian filtering), where the images act as noisy observations of the reality. The nature of inference methods allows abstracting from the specific problem or scenario and modeling very different situations under the same framework. In this field, particle filters (or non-linear, non-Gaussian filters) have emerged as a popular tool to make estimates of posterior distributions within the Bayesian frame. In this work, we review the use of particle filters and exemplify its performance with two typical vision problems: human tracking, and vehicle tracking.
BIB_text
@Article {
author = {Marcos Nieto and Oihana Otaegui},
title = {Probabilistic object tracking for global optimization},
abstract = {
Object tracking in video-surveillance applications is a major topic in the signal processing and computer vision community. Its aim is to estimate properties of the imaged objects within a scene, such as their dimensions, motion, visual features, etc. Typically, object tracking is treated as an inference problem through probabilistic methods (Bayesian filtering), where the images act as noisy observations of the reality. The nature of inference methods allows abstracting from the specific problem or scenario and modeling very different situations under the same framework. In this field, particle filters (or non-linear, non-Gaussian filters) have emerged as a popular tool to make estimates of posterior distributions within the Bayesian frame. In this work, we review the use of particle filters and exemplify its performance with two typical vision problems: human tracking, and vehicle tracking.
}
date = {2010-11-16},
year = {2010},
}
author = {Marcos Nieto and Oihana Otaegui},
title = {Probabilistic object tracking for global optimization},
abstract = {
Object tracking in video-surveillance applications is a major topic in the signal processing and computer vision community. Its aim is to estimate properties of the imaged objects within a scene, such as their dimensions, motion, visual features, etc. Typically, object tracking is treated as an inference problem through probabilistic methods (Bayesian filtering), where the images act as noisy observations of the reality. The nature of inference methods allows abstracting from the specific problem or scenario and modeling very different situations under the same framework. In this field, particle filters (or non-linear, non-Gaussian filters) have emerged as a popular tool to make estimates of posterior distributions within the Bayesian frame. In this work, we review the use of particle filters and exemplify its performance with two typical vision problems: human tracking, and vehicle tracking.
}
date = {2010-11-16},
year = {2010},
}
