Simulation And Prediction Of Pedestrian Movement Based On A Biped Model And A Microscopic Evaluation Method

Since the beginning of the 21st century,the people-oriented development concept has gradually penetrated the hearts of the people.How to provide people with a safe,comfortable,convenient and colorful living environment is one of the important di-rections for scientists to work hard.Research,simulation and prediction of pedestrian movement are widely used in safety engineering,architecture and transportation facili-ties design,games and animation production,robot navigation,automatic driving,etc.,and are one of the ways to build a people-oriented social environment.However,in the existing research,the research on limb movement characteristics and decision-making of movement is mostly independent.For example,when simulating a pedestrian group movement or a single trajectory,the influence of limb movement is usually not con-sidered,and the pedestrian is regarded as a particle.This approximation simplifies the simulation to a certain extent,but at the same time ignores the impact of some micro-scopic step level characteristics,such as the periodicity of the biped movement,the dynamic adjustment of the step size,the change of the occupied volume,the limita-tion of the joint angle,and the correlation between the steps,etc,on the macroscopic movement.In this paper,we propose a biped model that takes into account the movement characteristics at the step level.By using the camera and the laser radar,the gaits of pedestrians at different speeds and different steering angles in controlled experiments are obtained.Based on the gait data,the step simulator,the step controller and the step predictor are developed to simulate and predict the pedestrian movement.Furthermore,in order to evaluate different pedestrian dynamic models in different scenarios,we de-veloped a micro-evaluation method based on trajectory level comparison.Based on this method,we optimized the parameters and evaluated the performance of a social force model,and the step controller and predictor in the biped model.At first,we build a a one-dimensional bipedal model to simulate the crowd move-ment of pedestrians in a single-lane scenario.The model consists of two parts,which are the one-dimensional stepping simulator to generate the position,orientation and time of the the next step with a speed when pedestrian walking along a line,and the one-dimensional step controller that the pedestrian plans his next step according to the walking state of the pedestrian in front of him.The step simulator outputs the step width,the step size,the external foot angle and the step cycle of the next step accord-ing to different speeds.The step controller first plans the approximate speed according to the head distance between adjacent pedestrians,and then determines the maximum speed that does not collide with the pedestrian in front,and finally outputs the speed of the next step after the acceleration limit.The simulation results show that the model can simulate pedestrian single-lane movement at very high density(up to 3 people/m).The simulated density-velocity relationship is consistent with the experimental ones,and can reproduce the stop-going wave with critical density.Also,the "lock-step phenomenon"is reproduced for the first time.Then we build a two-dimensional biped model to simulate and predict the footprint of a single pedestrian in a continuous-targets scenario.The model consists of three parts,which are the two-dimensional step simulator that calculates the position,orientation and time of the next step based on the steering angle.The step controller that generates the next steering angle according to the relationship between the current movement state and the targets,and a step predictor that predicts the position of the next three steps based on the observed steering angle of the current step.The basis for the establishment of the two-dimensional biped model is the controlled steering experiment of a single pedestrian.Based on this experiment,we fit the relationship between single step length,single step length,ankle torsion angle and turn angle as the core of the step simulator.The core of the step controller is the change of the turn angle in the steering experiment with different angles.With further observation on the variation of the turn angle in the experiment,we propose a "four-step hypothesis":the pedestrian's single turn process consists of up to four steps,and the change of turn angles in the four steps is correlated.Based on this assumption,we extract the frequency distribution of the turn angles of the last few steps when the turn angles of the first several steps are in different intervals as the core to drive the step predictor.The simulation results show that the step simulator can accurately simulate the footprints with the known turn angles.The step controller can accurately output the turn angle value during the whole movement with predefined targets.The step predictor is able to predict the next three steps accurately according to the detected turn angle of the current step.At last,we propose a microscopic-movement-data based microscopic evaluation method to evaluate a pedestrian dynamic model.The core idea of the method is to bring the model into the pedestrian in the experimental data in the first person.Then the model drives the movement of the pedestrian in the next period of time,while other pedestrians still move according to the data recorded in the experiment.At the end of this period,we can draw qualitative conclusions by counting and visualizing the directional error which is obtained by comparing the experiment and simulation results.Normalize the directional error,we can get quantitative evaluation results with considering errors in different aspects.Based on this method,we evaluate and optimize the parameters of a social force model and the two-dimensional biped model.The results show that the qualitative conclusions obtained by this method are consistent with the conclusions ob-tained by a macroscopic method,and the models with smaller error in the macroscopic method tend to be better at the macro level.The one-dimensional biped model proposed in this paper can help crowd manage-ment and provide a model basis for simulating falls and trampling.The simulation part of the two-dimensional biped model provides a new way to simulate human motion in games and animations,and prediction part can provide new ideas for robot navigation and autonomous driving.At the same time,it also lays the foundation for the develop-ment of two-dimensional biped model for crowd movement.Finally,the microscopic evaluation method provides a universal and effective method for the verifying of dif-ferent pedestrian dynamic models in different scenarios.