Pedestrian Trajectory Prediction By Using Scenes And Pedestrian Forces

With the development of autonomous driving technology,people pay more attention to personal safety.In order to give drivers more sufficient reaction time to deal with emergencies,scholars began to study the path of pedestrians and put forward the pedestrian path prediction algorithm based on this.Because of the strong subjectivity of the pedestrian track,it is full of uncertainty in the prediction.Therefore,under the influence of a variety of adverse factors,the analysis of pedestrian movement,pedestrian state prediction and other issues become complex.To sum up,the study of pedestrian path prediction has high theoretical value and practical significance.In this paper,the pedestrian path prediction method based on the interaction between scene and pedestrian is studied.By classifying the factors that affect the pedestrian movement path and extracting the features respectively,the accuracy of the prediction path is increased.The following improvements are proposed,including the following main contents: 1.In order to accurately express the influence between pedestrians,a neighborhood weight expression based on Mahalanobis distance was proposed.In the graph convolutional neural network,the value in the adjacency matrix represents the edge in the graph,and in this paper,it represents the interaction between pedestrians.In order to improve the prediction accuracy,this paper uses the Maobanobis distance formula to calculate the spatial distance,and converts it into the neighborhood weight,which is stored in the adjacency matrix,as the feature of pedestrian interaction.2.To solve the problem of interaction between pedestrians and scene,the pedestrian scene graph is put forward and set up the pedestrian neighborhood scene.To reflect the influence of static scene for pedestrians,pedestrian scene graph is proposed in this paper,which based on scene in a static environment modeling to construct the obstacle factors,and set up corresponding repulsion,the size of the static scene information,said the scenario at the same time for each pedestrians in scene graph set up neighborhood pedestrians scene,and press it with the distance between the pedestrian to update of resistance,with pedestrians scene neighborhood characterization of pedestrians scene graph,reflect the interaction between pedestrians and scenarios.3.In order to generate the trajectory with scene information and pedestrian interaction information,the interaction features of scene neighborhood and pedestrian were combined,and the temporal extrapolation convolutional network processing was used for data processing,and the pedestrian prediction trajectory was generated.In order to verify the effectiveness of the proposed model,this paper conducts experiments on two common data sets(ETH and UCY),and uses ADE and FDE as performance evaluation criteria.Compared with the previous proposed method,it is found that the proposed algorithm improves the prediction accuracy on the basis of reducing parameters.