Pedestrian Trajectory Prediction Based On Deep Learning

With the rapid development of automobile intelligence,unmanned driving technology has become a research hotspot in today ’s society.Pedestrian trajectory prediction is an important research direction in unmanned driving technology.Pedestrian trajectory prediction can prepare for planning safe driving routes and taking corresponding safety measures to avoid unnecessary traffic accidents.The trajectory of pedestrians is not only affected by the surrounding environment,but also by the social interaction between adjacent pedestrians,and the movement of individuals is different during the movement.How to accurately predict the trajectory of pedestrians is a very challenging problem.The traditional trajectory prediction method extracts pedestrian features manually and designs functions to represent the interaction between pedestrians.This method has low prediction accuracy in complex environments,and the generalization ability and robustness of the model are poor.The deep learning method can effectively make up for the problems of poor generalization ability and poor prediction accuracy in traditional models.Based on the above considerations,this paper studies the pedestrian trajectory prediction task,and proposes two trajectory prediction methods based on deep learning.The specific work is as follows:(1)Aiming at the problem that the Social-STGCNN algorithm has low prediction accuracy and ignores the blind area of pedestrian vision in trajectory prediction,a pedestrian trajectory prediction algorithm based on spatio-temporal graph convolution is proposed.Firstly,the spatio-temporal interaction information of pedestrians is obtained through the spatiotemporal graph,and the connection between unrelated points in the graph network is removed according to the interaction information.Then,the interactive features in the spatio-temporal graph are extracted by graph convolution.Finally,the time extrapolation convolutional neural network(TXP-CNN)is used to predict the future trajectory of pedestrians.The experimental results on the public datasets(ETH and UCY)show that the average displacement error(ADE)and final displacement error(FDE)of the improved algorithm model on this dataset are improved by 4.5 % and2.7 % respectively compared with the previous algorithm.(2)Aiming at the interaction between pedestrians and scenes and the interaction between pedestrians in pedestrian trajectory prediction,this paper proposes a graph convolution pedestrian trajectory prediction algorithm based on attention mechanism.Firstly,VGG19 is used to extract scene information,and the interaction information between pedestrians and scenes is obtained through the physical attention mechanism module.Secondly,the social attention mechanism is used to obtain the social interaction information between pedestrians.Then,the interaction information between pedestrians and scenes and the pedestrian interaction information are aggregated and input into the graph convolution and feature extraction.Finally,the time extrapolation convolutional neural network is used to predict the future trajectory of pedestrians.The experimental results on public datasets(ETH and UCY)show that the improved algorithm model has improved trajectory prediction performance.