Research On Trajectory Generation Strategy Based On Moving Target Capture

As self-driving cars,social robots and intelligent transportation systems become an essential part of everyday life,the prediction of the future trajectory of pedestrians becomes an urgent need.Thus,a specific task is to predict the most likely future trajectory given a dynamic scenario(for example,an intersection)based on the observed pedestrian trajectory.However,due to the complexity of human behavior,it is challenging to accurately predict the future path of pedestrians.In recent years,with the continuous development of the field of artificial intelligence and the vigorous development of data-driven technology,the research prospect of pedestrian trajectory prediction has become increasingly clear.The precondition to improve the accuracy of pedestrian trajectory prediction is to model its interaction.In this paper,the existing trajectory prediction model is studied and analyzed,and an improvement scheme is proposed to address S-GAN’s shortcomings in scene information extraction and pedestrian interaction.A prediction model is built by combining the attention pooling mechanism with the generation of adduction network,which is named A-CGAN.First,the potential features are extracted by setting the pedestrian feature extraction module and scene information extraction module,and the influence weight is allocated by introducing the attention pooling mechanism to pedestrians in the scene and objects that may affect pedestrian tracks in the scene,so that the model can make full use of interactive information.Secondly,based on the idea of SFM,the social interaction of target pedestrians is simulated by allocating the weight of corresponding pedestrian and scene information,and random sampling noise is added to simulate the disturbance force caused by the complexity of target pedestrians’ own behavior.Thirdly,the adversarial loss is introduced in the process of model training to solve the problem of multi-mode trajectory prediction.The existing pedestrian dataset cannot provide path data of multiple real tracks,and human behavior is characterized by diversity and flexibility.This paper uses carla to create a multi-track simulation dataset,which not only reproduces the real scenes under the classical data set,but also creates two additional interaction scenes,and marks several tracks in line with social behavior norms for pedestrians with possible multi-track paths.This can enrich the training data and improve the generalization ability of the model.The trained model is tested on the classical data set and the multi-trajectory simulation data set created in this paper.By comparing with the classical model,the effectiveness and superiority of the prediction performance of the proposed model are verified.