Robotics Research Lab
CRES
USC Computer Science
USC Engineering
USC
/ Research / Projects / NSF HSD

Overview Approach Images and Videos
Publications Support Contact Details

Top
Overview

University of Southern California (USC), USC Information Sciences Institute (ISI) and Brandeis University researchers are collaborating on an HSD program funded project that explores the nature of individual and collective spatial dynamics. This systematic and coordinated multidisciplinary effort brings together expertise in computational sciences, mathematical modeling, and cognitive science disciplines. The project aims to understand how individuals and groups explore complex spaces, how social cues affect collective behavior, and how a mathematical link can be established between microscopic and macroscopic models of adaptive human behavior. Anonymous data captured from real world group interactions within public spaces, made possible through collaboration with the Los Angeles California Science Center and Boston Museum of Fine Arts, show spatio-temporal structure within collective pedestrian and sight-seeing behavior. Algorithms currently being developed will enable automated processing of this data which will make statistical analysis of long-term patterns feasible. A macroscopic model of crowd flow through a museum exhibit, also currently under development, uses the captured data to verify modeling assumptions and to establish empirical values for open parameters. This model already provides insights towards improved exhibit design and suggests possible efficiency differences between uni- and bi-directional crowd flows. The scanning laser and video data were captured in locations with high crowd density and significant traffic flow. This is an important data set as it has direct application to studies of evacuation and emergency egress scenarios.

While these data and models are significant in themselves, the project also considers the role of peoples' cognitive representations of space in a social setting. Thus, the team has begun modeling of a three dimensional digital version of the California Science Center which will allow human participants to navigate exhibits in controlled conditions. Unlike existing work, this museum experience will include digitally animated individuals and crowds that explore and utilize space in a manner that is macroscopically consistent with real world data. This project points the way forward to what is potentially a new research methodology. Such a methodology is important for science because it is increasingly recognized that important phenomena may be overlooked by reductionist approaches (e.g., behavior rooted in interaction dynamics). The HSD program's cross-cutting nature allows exploration of such higher-order phenomena, bringing together necessary expertise, and transforming methods so that critical questions can be taken up.

Approach

We have captured video data from cameras overlooking open spaces within the museum in order to complement the data collected from the life-tunnel exhibits.

We use an automated procedure to detect people within the frames. The low-resolution in the input video means that people are only about 15 pixels across. Our implementation looks for circular regions of high-contrast, as detailed below.

The objective is to calculate a spatial density function over the monitored space.

Our implementation maintains a model of the image background and slow changes due to lighting changes. A Generalized Hough Transform is applied to pixels which differ sufficiently from the background in order to find circles with suitable radii.

Images and Videos

The following images typical performance of the software used to detect the presence of humans in video frames. The images are frames from a video capture at a McDonald's in the California Science Center. As can be seen, the video frames provide no information about the identities of the visitors within the museum.

Tracker Sample 1


Tracker Sample 2

Publications

There have been no publications as of yet.

Support

This project is funded by the National Science Foundation's Human and Social Dynamics program.

Contact

Danko Krajisnik (krajisni at usc dot edu)
Dylan Shell