Efficient Computational Models for Simulating Large-Scale Heterogeneous Crowds
Abstract
Understanding the behavior of pedestrians in a crowded scene has many real-world applications. It has been the subject of extensive research in multiple domains, including applied mathematics, robotics, psychology, sociology, civil and traffic engineering, architectural and urban design, etc. Many applications such as training for battlefield simulation and urban warfare, intelligent surveillance, management of large mobs or unruly crowds, as well as use of robots in battlefields and dangerous environments need improved capabilities to simulate large crowds. Such crowds are characterized based on number of agents or pedestrians (e.g. large crowds with tens or hundreds of thousands of people), high densities, as well as heterogeneous or varying behaviors. Current state of the art is not able to model such large and diverse crowds that arise in different applications. Motivated by the practical demands of modeling and simulation and better understanding of dynamic aggregate behaviors that are observed in modern-day Megacities, we propose to develop novel computational models and real-time crowd simulation algorithms that can be used to for battlefield simulation, personnel training, and design evaluation. More specifically, we propose to address the following problems with innovative approaches: Accurate pedestrian geometrical models based on elliptical shapes and bio-mechanical constraints; Development of novel local navigation algorithms, including collision avoidance and orientation computation; Modeling of intricate pedestrian dynamics on uneven terrains; Interactive simulation and visualization of crowd flows using a continuum formulation; Innovative hybrid combination of discrete-continuum crowd simulation and statistical prediction; Novel data-driven crowd simulation algorithms using Bayesian learning that exploit the behaviors captured using videos and sensor data; Real-time simulation algorithms that exploit the parallel capabilities of multi-core CPUs and GPUs; Applications of these algorithms for architectural and urban design, virtual-reality training, planning of emergency response, as well as collision-free navigation of robots between pedestrians. Scientific Merits: This research is expected to lay the scientific foundation for addressing crowd simulation and societal-scale problems related to large-scale crowd behaviors and movements. It will introduce new algorithmic developments, efficient computational methodologies, parallelization, and system integration for several compelling applications and evaluation of these approaches. We expect to (1) attain plausible explanations of the behavior of individual agents (e.g. pedestrians or vehicles) and how they interact with each other under various settings; (2) hypothesize and verify how heterogeneous agents interact across varying scales (e.g. a group of tens, hundreds, thousands, to millions) and across levels of social organizations, from individuals to groups; (3) develop new set of representations, algorithms, and systems that can be used for interactive crowd simulation in different important applications. ARMY/ DoD Relevance: The proposed crowd simulation framework based on both discrete and continuum formulations can provide computational tools for diverse applications, where modeling of realistic crowds and vehicles is much needed. Examples range from urban planning, autonomous vehicle system design, training of soldiers and marines in coping with large-scale crowds, etc. Moreover, the techniques and algorithms for modeling individuals and groups of individuals can be integrated with computer vision techniques for crowd surveillance, predictive tracking and monitoring in identifying possible terrorist suspects...
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 14, 2019
- Source ID
- W911NF1910069
Entities
People
- Ming C. Lin
Organizations
- Army Contracting Command
- United States Army
- University of Maryland