Modeling Crowds in Urban Spaces
1 like78 views
The document discusses the work of Javier Espinosa and colleagues on modeling crowds in urban spaces using big data and advanced visualization techniques. It outlines a general approach for tracking and analyzing crowd behavior in real time, identifying suspicious behaviors, and predicting responses to events. The challenges include integrating data analytics with realistic simulations and leveraging high-performance computing infrastructure for effective crowd management and security strategies.
Modeling Crowds in Urban Spaces
- 1. Javier Espinosa Postdoctoral Research Fellow, BSC-LAFMIA javier.espinosa@bsc.es MODELING CROWDS IN URBAN SPACES From Big Data to Smart Secure Regions ISUM, Puebla, 11-15th April, 2016 Research and Genoveva Vargas-Solar, Hugo Perez, Isaac Rudomin, J.L. Zechinelli-Martini
- 4. MIGRATION FLOW IN EUROPE
- 5. MIGRATION FLOW IN EUROPE Migratio n of Children to Europe
- 6. source: http://missingmigrants.iom.int MIGRANT ROUTES IN AMERICA, 2016
- 7. source: http://missingmigrants.iom.int MIGRANT ROUTES IN AMERICA, 2016 CONCERNS (PROJECT) Data Management & Visualization techniques for observing, simulating and analysing crowds in real time and helping in the development of security strategies
- 8. GENERAL APPROACH Location of individuals in a 2D space Computing and aggregating trajectories (behavior model) Observe individuals in- situs and at different granularities TRACKING1 Model & reproduce crowd behavior Learn from collected data Predict behavior as result of specific events Model updated continuously PROFILING2 Identify individuals not belonging to the crowd Detect suspicious behaviors ANALYSIS3
- 9. COMPUTATIONAL PROBLEM “GREEDY DATA PROCESSING” Iterative data processing and visualization tasks need to share CPU cycles Data is a bottleneck APPLICATION DRAM DISK/DATABASE CPU Multiples Cores GPU Thousands of Cores 1-5GBps1-10GBps
- 10. COMPUTATIONAL PROBLEM “GREEDY DATA PROCESSING” Data processing and visualization tasks need to share CPU cycles Iterative process Data is a bottleneck APPLICATION DRAM DISK/DATABASE CPU Multiples Cores GPU Thousands of Cores 1-10GBps1-10GBps CHALLENGES • Integrating computationally expensive data analytics with realistic character visualization within realistic environments • Giving the impression of reality to data analysers who will be able to make critical decisions
- 11. Realistic populated environments Rendering, visual variety, character animation, artificial intelligence and motion planning Parallel approaches for achieving real-time crowd simulations Understanding users’ mobility from data collections Emergency management and disaster response, trend Learning techniques Statistical models for predicting users location based on large datasets Knowledge based techniques Modelled and predicted trajectories, locations, and users and crowd preferences, Mined the correlation between users and locations in terms of user-generated GPS trajectories RELATED WORK Visualization Geo-Location
- 12. OBJECTIVES 1. Visualize in real-time the behaviour of individuals and groups moving and evolving within real environments, at different precision levels and granularities 2. Predict individuals’ and crowd behaviour in public urban spaces using information harvested from omnipresent surveillance 12
- 13. PROPOSED ARCHITECTURE NoSQL BLOBs Data Storage Data Analysis Crowd Stream Services Data Collection images Crowd DRONES Web App GPUs Crowd Simulation (3D Map) Simulation & Learning
- 14. Crowd Data Collection GeoLife GPS trajectory dataset 182 users, 17,621 trajectories 1.2 million Km. 48,000+ hours http://research.microsoft.com/en-us/projects/geolife/ CURRENT WORK: URBAN DATA
- 15. NoSQL LOBs Data Storage Data Analysis CURRENT WORK: ANALYSIS OF TRAJECTORIES NoSQL BLOBs Data Storage Data Analysis for t in user.trajectories: for p in t.points: drawPoint(p)
- 16. CURRENT WORK: 3D GEO VISUALIZATION Web App GPUs 3D Map Simulation Open-source JavaScript library for 3D globes and maps Custom software for displaying multiple animated pedestrians efficiently
- 18. CONCLUSIONS & PERSPECTIVES Conclusions Combining geo-tagged data with online data for visualizing crowds at different levels of precision and detail. Our data processing and simulation are computationally expensive and critical thus we relay on HPC infrastructure with hybrid architecture (CPUs + GPUs) to produce an efficient solution. Perspectives Harvest and process data produced by DRONES and security cameras Predict individuals and crowd behaviour Detect abnormal situations in presence of specific events 18
Editor's Notes
- #2: Join effor between France, Spain and Mexico combining expertise from different domains.
- #5: Exhodus are related to war crises
- #6: Exhodus are related to war crises
- #7: But exodus are also related to economic crises.
- #8: Since we concerned we proposed a project combining our expertises for helping having tools for dealing with these kind of situations.
- #9: Exhodus are related to war crises
- #10: impact on realistic rendering of realistic 3D environments populated with crowds
- #11: impact on realistic rendering of realistic 3D environments populated with crowds