Research on pedestrian trajectory prediction by GAN model based on LSTM

Abstract

As the application scenarios of unmanned vehicles are gradually diversified and complicated, how to improve the safety and efficiency of unmanned vehicles requires the prediction of dynamic obstacles for mobile robots, and this paper studies a pedestrian trajectory prediction algorithm that can be applied with unmanned vehicles in the first-view video images. A pedestrian trajectory prediction model method based on GAN model of LSTM is proposed. Firstly, in the feature extraction module, the observer camera motion state vector, pedestrian pose information and pedestrian history trajectory are considered as loses. Secondly, in the attention module, the motion attention mechanism is used to measure the degree of influence of the observer camera motion on the pedestrian trajectory, the pose attention mechanism is used to extract hidden features from the human pose, and the interaction attention mechanism is used to model the social interaction between pedestrians, and finally the GAN network module is used to obtain the predicted trajectory using generative adversarial networks. In this paper, we use ETH this dataset for model analysis, and the experimental results show that the proposed model of GAN model based on LSTM can effectively predict the future motion trajectory of pedestrians.