Modelo para a predição de comportamento de multidão em tempos futuros

• Autor(a) principal: Bianco, Clicéres Mack Dal
• Data de Publicação: 2018
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da PUC_RS
• Texto Completo: http://tede2.pucrs.br/tede2/handle/tede/8411
• topic: Multidão; Simulação; Predição; Crowd Simulation; Prediction; Pedestrian Dead Reckoning; CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO

The behavior of human crowds has been studied in several areas of knowledge, such as Human Sciences, Engineering and Computer Science. In the field of Computer Science, crowd simulations generate information that serves as the basis for the development of research with applications focused on security and entertainment. The objective of this work is to provide a computational model that estimates the behavior of the crowd at a certain future time. Methods such as Pedestrian Dead Reckoning (PDR) allow an approximate prediction of people’s positions in future times, using Physical hypotheses. However, the challenge is to generate accurate positions by considering realistic behavior of crowds in complex environments, where the estimation using only the physics of pedestrian movements may not be robust enough. This paper proposes a model called Time Machine (TM) that extends the concept of PDR with data resulting from interactions among people and complexity of the environment. This model has been integrated with BioCrowds, a crowd simulator that discretizes the available areas of the environment and represents them through marker (points in space) that allow the free movement of virtual humans. In order to validate the proposed methodology, several simulations were performed including case studies with real situations. For these case studies, analyzed error includes information on density and variation of pedestrian positions. An example is the prediction of 8.33 seconds in the future time of a simulated population of 320 agents in a free area of 338m2 , presenting the error of 0.25 in average. In addition, for cases with real data, the TM method estimated the crowd with the maximum error of 0.24.