Research On Pedestrian Trajectory Prediction Based On Generative Adversarial Networks

With the development of the age and the advances in technology,the Internet and all kinds of terminals have become an indispensable part of people's life.The popularization of sensors has generated a large amount of idle data in the society.Based on these data,the motion trajectory prediction of moving objects is of great value in social prediction,automatic driving and other fields.This paper mainly studies pedestrian trajectory prediction.Firstly,the conception of trajectory prediction and existing pedestrian trajectory prediction methods are systematically studied and analyzed.Then,two prediction methods are proposed based on the generative adversarial networks:In order to solve the problem that the pedestrian interaction features of the social generative adversarial networks are simple and the pedestrian interaction information cannot be fully utilized,a pedestrian trajectory prediction model based on social features and attention mechanism is proposed to model pedestrian interaction patterns more effectively.Model structure adopted generative adversarial networks.The interaction information of pedestrians is enriched by setting three social features,and the influence distribution of pedestrians in the same scene is carried out by attention mechanism,so that the network can make full use of the interaction information of pedestrians and improve the accuracy of the model.A pedestrian trajectory prediction model based on multi-object tracking and scene information is proposed.The pedestrian trajectory information is extracted from the video by multi-object tracking,and the model effect is improved by using the scene information and pedestrian interaction information.In this model,the generative adversarial network is model structure.The scene image information is obtained by video,so that the pedestrian-scene interaction information can be introduced.The social attention mechanism and scene attention mechanism are set up to allocate the weight of the pedestrian trajectory information vector,so that the network can make full use of the scene information and pedestrian interactioninformation,and improve the model accuracy.