标题：Motion imagery datasets capture evolving phenomena like the movement of a car or the progress of a natural disaster at video or quasi-video rates. The identification of individual spatiotemporal trajectories from such datasets is farm from trivial when these trajectories intersect in space, time, or attributes. In this paper we present our approach to this problem and relevant algorithms. A key component of our work is the attribute classification (AtC) strategy, a novel approach to classify individual trajectories using a sequence of image processing and neural network tools. Geometry, k-means clustering, backpropagation and self-organizing maps are the tools applied towards the classification of such datasets. Other key components of our approach include the novel g-SOM approach to generalize spatiotemporal datasets, and the concept of spatiotemporal helixes, used to model the behavior of individual objects. In this paper we present these key components of our approach and some experimental results
期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2003
卷号：XXXIV-5/W10
出版社：Copernicus Publications
摘要：Motion imagery datasets capture evolving phenomena like the movement of a car or the progress of a natural disaster at video or quasi-video rates. The identification of individual spatiotemporal trajectories from such datasets is farm from trivial when these trajectories intersect in space, time, or attributes. In this paper we present our approach to this problem and relevant algorithms. A key component of our work is the attribute classification (AtC) strategy, a novel approach to classify individual trajectories using a sequence of image processing and neural network tools. Geometry, k-means clustering, backpropagation and self-organizing maps are the tools applied towards the classification of such datasets. Other key components of our approach include the novel g-SOM approach to generalize spatiotemporal datasets, and the concept of spatiotemporal helixes, used to model the behavior of individual objects. In this paper we present these key components of our approach and some experimental results
