Please use this identifier to cite or link to this item: https://hdl.handle.net/11681/34755
Title: Robust methods for fusing heterogeneous spatiotemporal data
Authors: Ellison, Charlotte L.
Roth, Zachary J.
Chen, Crystal
Keywords: Geospatial data
Military intelligence
Tensor algebra
Publisher: Geospatial Research Laboratory (U.S.)
Engineer Research and Development Center (U.S.)
Series/Report no.: Technical Note (Engineer Research and Development Center (U.S.)) ; no. ERDC/TN GRL-19-3
Abstract: Modern technology has offered the capability to document and explore characteristics of human patterns of life. One such case is the abundance of time-indexed, ordered locations known as spatiotemporal trajectories. This increase in geospatial data, as well as other relevant contextual information, poses both opportunities and challenges to the Army geospatial community. This data has the potential to allow for gaining novel insight into human behavior, or opposite, previously developed methods may not be adequate for addressing the volume, sparsity, or complexity inherent in these data sets. New frameworks and methods for understanding the structure within the data will be key to providing additional insights into human behavior, which in turn, can expand the Army geo-intelligence-human-intelligence spectrum of operations. The analysis of an ST trajectory in isolation (i.e., in conjunction with no additional data) is able to identify only those characteristics that are represented in the trajectory information such as locations of interest, sequences of locations, and movement characteristics. Current methods of trajectory analysis can identify historical trends in locations or movements, but these methods lack the ability to detect interactions between human movement and the surrounding environment. In order to move beyond this limitation, data from multiple ST modalities, global positioning system signals, social media interactions, recordings of events, and more, must be analyzed in unison. Each such ST modality captures distinct features that have the potential to reveal key insights during an analysis. This work investigates the challenge of merging spatial trajectories with contextual features in a generalizable framework in order to understand the relationships between human movement and the surrounding environment. To do this, disparate ST datasets are merged together using a tensor-based representation that lends itself to higher dimensional data modeling and information extraction. This framework of data representation stores additional contextual information along with the spatial trajectories in question to allow for the detection of patterns that do not exist in the spatial trajectories alone. In brief, this paper (1) proposes a method of modeling spatial trajectories, social interactions, events, and more via binary tensors, and (2) applies a tensor decomposition to detect key features and discover hidden correlations within the data. This new framework will allow for greater understanding of human movement by revealing latent features in the data.
Description: Technical Note
Gov't Doc #: ERDC/GRL TN-19-3
Rights: Approved for Public Release; Distribution is Unlimited
URI: https://hdl.handle.net/11681/34755
http://dx.doi.org/10.21079/11681/34755
Appears in Collections:Technical Note

Files in This Item:
File Description SizeFormat 
ERDC-GRL TN-19-3.pdf862.46 kBAdobe PDFThumbnail
View/Open