Evaluation of Pedestrian Data Collection Methods Within a Simulation Framework



Evaluation of Pedestrian Data Collection Methods Within a Simulation Framework

Authors

M Fetiarison, G Floetteroed, M Bierlaire.Ecole Polytechnique Fédérale de Lausanne, CH

Description

This article describes a simulation framework that facilitates the design of pedestrian data collection campaigns. Major components of the simulations system and its application are presented.

Abstract

The estimation and validation of pedestrian behavioral models requires large amounts of detailed and appropriate data, the collection of which is a costly and time-consuming undertaking. The identification and design of an appropriate data collection method therefore is of great importance, which, however, is an arduous and itself time-consuming task. This article describes a simulation framework that facilitates the design of pedestrian data collection campaigns. The resulting software laboratory is designed to minimize (the preparation of) real data collection efforts.

Of course, a simulation laboratory alone is insufficient to generate the data needed to estimate a pedestrian model, and real data is needed at some point in the model estimation process. However, simulated data is very useful for the evaluation of a particular data collection campaign's efficiency in terms of cost, complexity, and ability to reveal the pedestrian behavior under consideration. The latter item is important because pedestrian behavior results from interactions with other pedestrians, which cannot be fully controlled in an experiment and is therefore hard to assess a priori.

The laboratory is built around a generic pedestrian simulator, which provides two major interfaces: The first interface links a pedestrian simulation model to the simulator that is used to generate what is subsequently called the "synthetic reality". The second interface extracts data from the synthetic reality that is gathered (equally simulated) sensors. This data is what would be costly to collect in a real experiment, but in a synthetic environment it is available in abundant amounts and at arbitrary quality. The following two paragraphs describe these interfaces in greater detail.

The simulator is designed in a generic way that takes into account different types of simulated models. In order to enable the simulation of many instances of a generic pedestrian model in a synthetical environment, the respective interface must take into account all possibly relevant information sources a (simulated) pedestrian may take into account when interacting with its environment. Also, the interface must enable the execution of all possible choice dimensions of a pedestrian. For the latter purpose we concentrate on the three-level framework of Hoogendorn et al. (2002), which distinguishes a strategic, a tactical, and an operational level. Phrasing this framework in terms of choice modeling, e.g. Bierlaire and Robin (2009), the lower level of this hierarchy corresponds to the choice of the next step, while higher levels comprise route and destination choice.

The interface through which the simulated data is provided also accounts generically for the wide variety of data that can be obtained from a pedestrian data collection campaign. On the one hand, this means that it must be possible to pass on the information provided by simulated sensors in the synthetic reality. On the other hand, it must also be possible to access information from the simulation that cannot be expected to be provided by existing sensors. Even if this data cannot be expected to be available in a real experiment, it may provide valuable information about processes in the simulation that reveal ways to improve the evaluated data collection campaign, for example the vision field of every pedestrian in the simulation.

Beyond these two major interfaces, the simulation system provides facilities for the specification of different scenarios. This includes the description of the physical environment, pedestrian characteristics, and the pedestrians plans (intentions). The latter essentially correspond to the behavioral instructions that are given to the individuals in a real experiment. For the representation of the physical environment, the widely used Drawing Interchange Format (DXF - Autodesk Drawing Interchange and File Formats, Release 14) is adopted.

The full article specifies the system in detail and demonstrates its application to the full design of a data collection campaign.

Publisher

Association for European Transport