Synthetizing Qualitative (Logical) Patterns for Pedestrian Simulation from Data

Abstract

This work introduces a (qualitative) data-driven framework to extract patterns of pedestrian behaviour and synthesize Agent-Based Models. The idea consists in obtaining a rule-based model of pedestrian behaviour by means of automated methods from data mining. In order to extract qualitative rules from data, a mathematical theory called Formal Concept Analysis (FCA) is used. FCA also provides tools for implicational reasoning, which facilitates the design of qualitative simulations from both, observations and other models of pedestrian mobility. The robustness of the method on a general agent-based setting of movable agents within a grid is shown.