Research On Pedestrian Trajectory Prediction Method Based On Generative Adversarial Network

With the development of the times and the progress of science and technology,the Internet and its various terminals have become an indispensable part of people’s life,and concepts such as smart city and Internet of things have also emerged.Due to the increasing number of intelligent autonomous systems in human society,the ability of these systems to perceive,understand and predict human behavior has become more and more important.Among them,human trajectory prediction based on the historical trajectory and environment of human motion can be widely used in city traffic,social robot,autonomous driving vehicle and other fields,so it has important research value.In recent years,the research on pedestrian trajectory prediction has increased sharply,but due to the variability of pedestrian movement,it has become a challenging task.In view of the shortcomings of most current research work lacking in considering many factors affecting pedestrian interaction,we carry out the research on pedestrian trajectory prediction method based on Generative Adversarial Network in this thesis.Firstly,the trajectory prediction problem and the existing pedestrian trajectory prediction methods are studied,and the advantages and disadvantages of traditional and deep learning trajectory prediction methods are compared and evaluated.Learning and obtaining the motion pattern of the target pedestrian based on Generative Adversarial Network(GAN),a GAN-based pedestrian trajectory prediction with deep learning framework is established,which provides a basis for subsequent model optimization and experimental design.Then,to tackle the interaction problem in pedestrian movement,a pedestrian trajectory prediction model based on Social-Interaction GAN(SIGAN)is proposed.A new Social-Interaction Module(SIM)is used to deal with the interaction between pedestrians,and the position and motion information between neighbor pedestrians and target pedestrians in a scene are considered.At the same time,a new measurement of spatial-temporal affinity is defined,which combines the location and velocity characteristics of all neighbor pedestrians.In addition,a local pooling mechanism is proposed,which can limit the interaction impact of neighbor pedestrians on the target pedestrian to an appropriate range.The experimental results show that SIGAN can achieve high accuracy.Furthermore,considering the motion intention information of neighbor pedestrians,a pedestrian trajectory prediction model called SR-LSTM and Velocity Attention based Social-Interaction GAN(SRVA-SIGAN)is proposed.SR-LSTM is used as location encoder to extract motion intention information effectively.At the same time,the influence of pedestrians in the same scene is reasonably assigned by setting the velocity attention mechanism.The experimental results show that SRVA-SIGAN model can further increase the accuracy of pedestrian trajectory prediction.The research results of this thesis can provide a new idea for the research of pedestrian trajectory prediction,and can also be used in practical applications such as unmanned driving,thus has good theoretical value and application prospect.