Real-Time Video Analysis of Pedestrians to Support Agent Simulation of People Behavior

Abstract

Online simulation coupled with real-time measurements of pedestrians in public buildings is a novel application which can be used to increase the security and safety of pedestrians within those buildings. To receive realistic forecasts it is necessary to update the simulation constantly against reality. The real-time video analysis can thereby support the simulation with the necessary data. This proposed system is largely supported by the state observer of the control theory.

Within a laboratory one room model pedestrian behavior is monitored and analyzed by video cameras. Evolving data is subsequently provided for the simulation where flow rates are additionally recorded. Thus, two resulting passenger flow rates are available which can be compared by the simulation controller. The actuators, receiving information from the controller, can afterwards trigger an appropriate action. Two different actuators have been realized in the model in order to bring the simulation passenger flow closer to the observed passenger rates: the velocity controller adjusts the walking speed of the passengers and the flow generator actuator has the ability to match the passenger generation rate.

Results show that the simulation passenger flow curve converges to the real passenger flow. As expected, the simulation curve follows the real passenger rate with a certain delay. Nevertheless, the simulation model appears to reflect the behavior of the persons in an appropriate way. Further investigations will show which additional instruments can be used to refine the simulation actor behavior.