Research On Simulation Method Of Crowd Evacuation Based On Reinforcement Learning And Deep Residual Network Learning

In recent years,the public emergencies or disasters with heavy casualties have occurred frequently.In a crowded area,the crowd crowding and trampling can easily be caused when an emergency occurs.To reduce the occurrence of dangerous events,it is necessary to conduct evacuation drill for the crowd in advance.The practical evacuated drills are expensive and cannot reflect the real behavior of the crowd in an emergency.Recent years,more and more attention has been paid to the computer simulation technology,which can be used to build a scene model and simulate the evacuation process of crowd,and then provide guidance for crowd evacuation in emergency.Therefore,it is of great social significance to use computer simulation technology to study crowd evacuation.Traditional methods of crowd evacuation mainly focus on improving crowd evacuation efficiency while many hypothetical rules or models reduce the realism of crowd evacuation.The data-driven method is an effective way to enhance the realism of crowd evacuation whose main idea is to approach the trajectories of crowd movements in video as much as possible.Machine learning has become a hot field,the combination of data-driven and machine learning methods has been proposed and applied in crowd evacuation.Both of these methods obtain the velocity or trajectory of crowd movement based on the video data.However,most of them are end-to-end learning methods,so they are highly dependent on data and have low flexibility.To address these problems,we propose a simulation method of crowd evacuation based on reinforcement learning and deep residual network learning.A more realistic and flexible simulation model is obtained by synthetically using the characteristics of crowd motion,machine learning algorithm and the calculation method of crowd motion simulation.This method belongs to non-end-to-end learning,which makes full use of the crowd motion characteristics and simulation algorithms,and reduces the dependence on data.The main work and innovation of this paper are summarized as follows:(1)Most of the existing evacuation studies lack the consideration of variable scenarios and dynamic motion.In this paper,we present a reinforcement learning based data-driven crowd evacuation(RL-DCE)method.First,a data-driven crowd evacuation(DCE)model is established.The model quantifies the crowd cohesiveness based on the characteristics of crowd movement in the video.Then,we propose a cohesiveness based K-means(C-K-means)algorithm to group the crowd and merge the individual's trajectories.Second,a hierarchical path planning mechanism is proposed.In this mechanism,the top-layer employs the obtained grouping trajectories and the Q-learning algorithm to plan the group path.The bottom-layer is mainly in charge of obtaining individual paths and using the reciprocal velocity obstacles(RVO)model to avoid collision.Finally,the realistic rendering method is deployed to obtain the crowd simulation results.The experimental results show that the proposed method can simulate the crowd evacuation more realistically in the dynamic environment.(2)In view of the lack of scene adaptability in current simulation research,we propose a deep residual network based crowd simulation method.First,a data-driven crowd properties quantization(DCPQ)mechanism is proposed.This mechanism quantifies social properties using the physical properties of the crowd.Second,a residual network for crowd behavior properties learning(ResNet-CBPL)motion prediction model is established.The model takes the crowd properties as parameters to construct the ResNet,and uses real data to learn the movement rules of the crowd.Finally,the realistic rendering method is deployed to obtain the crowd simulation results.The experimental results show that the proposed method can simulate the crowd motion more realistically.The method has higher flexibility and can be applied to various scenes.(3)A three-dimensional simulation platform based on real data is established.The platform is based on Microsoft Visual Studio 2013 and XNA 4.0.The crowd evacuation simulation results are studied and analyzed in scenes of Jinan Quancheng Square,campus road,office and so on.The experimental results show that the proposed method can simulate the evacuation movement of crowd more realistically,and the simulation results are consistent with the real scene.