In a large body of reports on simulations of pedestrian movements, pedestrians are often assumed to walk along the shortest path. However, they are likely to choose different routes according to their own preferences. In this paper, we propose some models which describe pedestrians' route choice behavior, by considering the influence of walking distance, the number of turns, street width, and landmarks. Furthermore, we discuss regional and/or individual characteristics of the route choice behavior of pedestrians.
First, we perform a survey of route choice by using the actual maps, which have different street shapes. This is because, we expect that pedestrians' route choice behavior will vary according to street shapes. We select three areas which have different street shapes. First region (Region 1) is composed of latticed road network, the second one (Region 2) is composed of complicated road network, the third one (Region 3) has some latticed road networks which are allocated in different direction.
Next we grasp the basic characteristics of pedestrians' route choice by simple statistical analysis. As a result, in Region 1 it is shown that pedestrians' route choice is influenced by road width. In Region 2 they are influenced by road width and landmarks, and in Region 3 the influence of the number of turns and landmarks is significant.
Second, we construct pedestrian route choice models by considering the influence of the above four factors. These models are based on the Dijkstra algorithm in which the distance function is composed of a liner function of the above four factors. The unknown parameter of each factor is called "influence coefficient". Then, to easily understand the relation between pedestrian route choice and influence coefficients, we produce a conceptual chart called “pattern cubes/ pattern charts”. Moreover, we analyze the relationships between the pedestrian route choice estimated by the models and the values of influence coefficients.
Third, we define “coincidence ratio” as an index of descriptiveness of and consistency of the proposed model. In Region 1, the location of turning points can not be estimated precisely by the model. However, in Region 2, the value of coincidence ratio is higher compared with Region 1. This is because the model can estimate many and various pedestrian route choice according to the values of influence coefficients. In Region 3, the number of estimated routes and the coincidence ratio are in between the results of Regions 1 and 2.
Finally, we classify responders of the survey into three types according to the values of the estimated influence coefficients; “pedestrians whose route choices are influenced by the number of turns and landmarks (Group A)”, “pedestrians who prefer the shorter route (Group C)”, and “pedestrians between Groups A and C (Group B)”. It is shown that pedestrians of Group A feel that they are likely to get lost, and feel difficulty in reading maps, and often drive cars. Especially, pedestrian who have difficulty in reading maps are most remarkable in Group A. Thus, we demonstrate that there is the strong relationships between the choice of simple route and the capability of reading maps.
