Search

Crowd simulations imitate the group dynamics of individuals in different environments. Applications in entertainment, security, and education require augmenting simulated crowds into videos of real people. In such cases, virtual agents should realistically interact with the environment and the people in the video. One component of this augmentation task is determining the navigable regions in the video. In this work, we utilize semantic segmentation and pedestrian detection to automatically locate and reconstruct the navigable regions of surveillance-like videos. We place the resulting flat mesh into our 3D crowd simulation environment to integrate virtual agents that navigate inside the video avoiding collision with real pedestrians and other virtual agents. We report the performance of our open-source system using real-life surveillance videos, based on the accuracy of the automatically determined navigable regions and camera configuration. We show that our system generates accurate navigable regions for realistic augmented crowd simulations.

A generative model is a statistical model that is able to generate new data instances from previously observed ones. In the context of business processes, a generative model creates new execution traces from a set of historical traces, also known as an event log. Two families of generative process simulation models have been developed in previous work: data-driven simulation models and deep learning models. Until now, these two approaches have evolved independently and their relative performance has not been studied. This paper fills this gap by empirically comparing a data-driven simulation technique with multiple deep learning techniques, which construct models are capable of generating execution traces with timestamped events. The study sheds light into the relative strengths of both approaches and raises the prospect of developing hybrid approaches that combine these strengths., Comment: 12 pages, 4 figures

One of the primary driving factors in building energy performance is occupant behavioral dynamics. As a result, the layout of building occupant workstations is likely to influence energy consumption. In this paper, we introduce methods for relating lighting zone energy to zone-level occupant dynamics, simulating energy consumption of a lighting system based on this relationship, and optimizing the layout of buildings through the use of both a clustering-based approach and a genetic algorithm in order to reduce energy consumption. We find in a case study that nonhomogeneous behavior (i.e., high diversity) among occupant schedules positively correlates with the energy consumption of a highly controllable lighting system. We additionally find through data-driven simulation that the na\"ive clustering-based optimization and the genetic algorithm (which makes use of the energy simulation engine) produce layouts that reduce energy consumption by roughly 5% compared to the existing layout of a real office space comprised of 165 occupants. Overall, this study demonstrates the merits of utilizing low-cost dynamic design of existing building layouts as a means to reduce energy usage. Our work provides an additional path to reach our sustainable energy goals in the built environment through new non-capital-intensive interventions., Comment: 25 pages, 10 figures

Augmenting virtual agents in real crowd videos is an important task for different applications from simulations of social environments to modeling abnormalities in crowd behavior. We propose a framework for this task, namely for augmenting virtual agents in real crowd videos. We utilize pedestrian detection and tracking algorithms to automatically locate the pedestrians in video frames and project them into our simulated environment, where the navigable area of the simulated environment is available as a navigation mesh. We represent the real pedestrians in the video as simple three-dimensional (3D) models in our simulation environment. 3D models representing real agents and the augmented virtual agents are simulated using local path planning coupled with a collision avoidance algorithm. The virtual agents augmented into the real video move plausibly without colliding with static and dynamic obstacles, including other virtual agents and the real pedestrians.

In this paper, we introduce a machine learning-based simulation framework of general-purpose multibody dynamics (MBD). The aim of the framework is to construct a well-trained meta-model of MBD systems, based on a deep neural network (DNN). Since the main advantage of the meta-model is the enhancement of computational efficiency in returning solutions, the modeling would be beneficial for solving highly complex MBD problems in a short time. Furthermore, for dynamics problems, not only the accuracy but also the smoothness in time of motion solutions, such as displacement, velocity, and acceleration, are essential aspects to consider. We analyze and discuss the influence of training data structures on both aspects of solutions. As a result of the introduced approach, the meta-model provides motion estimation of system dynamics without solving an analytical equation of motion or a numerical solver. Numerical tests demonstrate the performance of the proposed meta-modeling for representing several MBD systems.