Research On Crowd Evacuation Method Based On Parameter Inversion Estimation And Dynamic Perception Modeling

In recent years,emergency security incidents have occurred frequently,causing serious losses to personnel and property,and posing a great threat to public safety and social stability.In order to reduce the losses caused by emergency events,we should preemptively grasp the characteristics and laws of individual movement and crowd evacuation.Crowd simulation technology is a technology that can simulate and match human motion behavior in different scenarios,with the characteristics of low cost,high repeatability,and high flexibility.However,the current crowd simulation models still have two major problems: 1.The simulation models often use default parameter configurations,resulting in large differences between simulation results and real data,and low accuracy.2.The characterization of real human behavior in the simulation model is not comprehensive enough,resulting in low realism.To solve the above problems,this paper proposes an effective approach to predict and evaluate pedestrian motion and crowd evacuation processes based on parameter inversion estimation and scene dynamic perception.The main contributions of this paper are as follows:1.A differentiable parameter estimator called ORCANet based on inversion is proposed to improve the accuracy of the simulation model.The estimator uses real video as input data and employs ORCA motion model and deep neural network to invert the parameters of pedestrians.To address the issue of non-differentiable operations involved in the ND parameters,this paper approximates the role of neighbors in avoidance with Gaussian kernel function,transforming the original discrete operation into differentiable operation.Utilizing the automatic backpropagation function of modern deep learning frameworks,the model can automatically invert the parameter combination of the simulation model and optimize the performance of pedestrian simulation models.The experimental results show that the model can converge to the correct parameter values on synthetic data and has been effectively validated on real-world datasets,generating parameter combinations that are closer to the actual trajectories of pedestrians and improving the accuracy of the simulation model.2.A dynamic perception model for intelligent agents in evacuation scenarios is proposed in this paper,which enhances the realism of the simulation model.Initially,the individual evacuation ability and knowledge are modeled for improving the agent’s perception in the evacuation scenario.Then,the stress system model is introduced,and the original General Adaptation Syndrome(GAS)model is improved.Finally,a dynamic perception model is proposed by combining the above three modules.This model integrates three modules: emotional contagion,knowledge dissemination,and stress response.Due to the introduction of stress response,the individual’s evacuation ability can be automatically adjusted by the influence of knowledge dissemination and emotional contagion.Based on the mastery level of evacuation knowledge,the individual can decide whether the stress response has a positive or negative impact,leading to changes in evacuation ability.A series of experiments and comparisons with real data have validated the effectiveness of this model in enhancing the simulation’s realism..3.A crowd evacuation simulation system for real-world fire scenarios is built.The system is constructed by the Unity engine and combines the two methods proposed in this paper,including four layers of architecture: user input layer,core algorithm layer,simulation rendering layer,and data processing layer.The system provides various visual analysis functions for evacuation simulation experimental results to assist in analyzing crowd evacuation situations.In addition,simulation experiments have validated the practicality of the system.