Direction-Decision Learning Based Pedestrian Flow Behavior Investigation

Zhe Zhang,Limin Jia

doi:10.1109/access.2020.2964001

Zhe Zhang, Limin Jia

Open Access

https://doi.org/10.1109/access.2020.2964001

Copy DOI

Abstract

To investigate the pedestrian flow behavior in corridors, a microscopic simulation model of pedestrian flow is proposed in this paper based on the desired-direction-decision learning and social force model. The proposed model is composed of two parts: direction-decision and walking behavior decision. First, the decision tree model is proposed to predict the walking direction of pedestrians by comparing the prediction and simulation performance of three different models. Then, to avoid collisions between pedestrians and obstacles, the acceleration model and the collision avoidance model are proposed to compute the walking speed. Finally, an computational experiment is conducted to simulate crowd movement in corridors. The experimental results show that the proposed model can suggest the shortest overtaking route for individual pedestrians among four models, and the speed-density relationship fits the experimental data well. The sensitive analysis shows that the lanes in bidirectional pedestrian flow can be formed much more easily if the pedestrians have higher direction changing frequency, and there is an optimal visibility field (2.8) to realize the highest traffic efficiency.

Highlights

The mathematical modeling of pedestrian flow has gained much scientific interest in recent decades because pedestrian movement is an important component in both fields of safety and capacity assessment of walking facilities [1], [2]
The most popular discrete models used in recent research include the lattice gas model (LGM) [4], [5] and the cellular automata model (CAM) [6]–[9], in which the space is discretized to approximate real pedestrian movement
We proposed a new simulation model of pedestrian flow by combining the advantages of the artificial intelligence method and social force model (SFM) in this paper

Summary

INTRODUCTION

The mathematical modeling of pedestrian flow has gained much scientific interest in recent decades because pedestrian movement is an important component in both fields of safety and capacity assessment of walking facilities [1], [2]. A linear model was proposed to determine the walking direction of pedestrians in real time [27]; the movement rules in the proposed method did not consider collision avoidance to improve simulation efficiency. The conflicts between pedestrians cannot be avoided if the predicted velocity is not sufficiently accurate, and the model performance relies much on the prediction performance To overcome this shortcoming, we proposed a new simulation model of pedestrian flow by combining the advantages of the artificial intelligence method and SFM in this paper. We proposed a new simulation model of pedestrian flow by combining the advantages of the artificial intelligence method and SFM in this paper Within this approach, the desired direction is produced by the prediction model. The social force model is used to solve the collision problem, which may not be solved by only the machine learning method

MODEL FRAMEWORK

ACCELERATION AND BODY COLLISION MODEL

DATA DESCRIPTION

MODEL PERFORMANCE COMPARISON

CONCLUSION AND EXTENSION