Home | | Book | | Research | | Publications | | Bio | | Press | Geosimulation Labs | |
Dr. Paul M. Torrens, Center for GIS, Department of Geographical Sciences, and UMIACS, University of Maryland |
Big data movement analytics
Project overview | Eye candy | Related groups | |
Project overview | |
Many location-aware devices now provide ongoing streams of movement data, often with allied action, activity, interaction, and transactional data following not too far behind. These silos of data enable a suite of geographical analysis and modeling techniques, which themselves then produce further data. This is particularly true of agent-based models tasked with representing fine-resolution characteristics of movement and interaction in massively dynamic systems. There is a need for smart data-mining and knowledge-building schemes that can work on these data, and all of the thorny complexity that implies. This project is focused on producing new techniques for extracting features, processes, and phenomena from movement data-sets generated by agent-based models. It builds on our earlier work on validating agent-based models using complexity signatures and space-time analyses, as well as our work on data-mining and machine-learning. The tools being developed are working in tandem with our suite for agent-based modeling. |
|
Eye candy | |
Geographical wisps: movement trails of synthetic walkers in simulation (lines), indexed by collision (beads) and time to collision (bead color from pink to red, high to low).
Relative speed of collective movement through a downtown streetscape
Individual wire movement tracks are extruded by ambient speed in the crowd flow
Collision beads atop space-time movement tracks (projected by relative ambient speed as above). The beads index collision potential in the vicinity of the track. In this case, a group of walkers were in motion side-by-side and following. Above, the illustration provides just a small window on the data. For 50 synthetic walkers on the small simulated streetscape, the model generates a total of 2.997 million collision-cast points per minute of simulated "real-time" interaction. In other words, at 30 checks-per-second, the walkers collectively build this mental map of their ambient surroundings.
Above, a view of a larger portion of the data-set. The persistent colored paths are actually strings of collision beads, and are illustrative of convoying of synthetic walkers in close proximity. |
|
Related groups | |
|
|
Robot motion control |
|
Human behavior in critical scenarios |
|
Modeling riots |
|
A toolkit for measuring sprawl
|
|
Simulating crowd behavior |
|
Wi-Fi geography |
|
Simulating sprawl |
|