A Pedestrian Trajectory Prediction Method Based on Generative Adversarial Networks

Abstract

The invention discloses a pedestrian trajectory prediction method, which mainly includes the following steps: preprocessing data and converting it into a matrix of [number of pedestrians, 4]; inputting the preprocessed data into an encoder to complete the process of generating a confrontation network. encoding function; input the output data after the encoder into the pooling layer to share the hidden information of the global pedestrian in the same scene; input the output vector after the hidden information pooling into the decoder to complete the generative adversarial network. The decoding function can initially obtain the predicted pedestrian trajectory data; the output vector of the generative adversarial network generator composed of the encoder, the pooling layer and the decoder is used as the input vector to enter the generative adversarial network discriminator for identification, and finally get the most Excellent model.

Description

Technical field [0001] The present invention relates to a pedestrian trajectory prediction method, based on generative adversarial network, suitable for predicting the future trajectory of pedestrians in complex scenarios. Background [0002] With the rapid progress of science and technology and the rise of artificial intelligence, automatic driving has gradually entered people's lives. In many first-tier cities at home and abroad, emerging things such as driverless buses and driverless school buses have gradually emerged. Predicting the future trajectory of current pedestrians can play a good auxiliary role in the application tools on unmanned platforms, such as driverless cars. This is because pedestrians and means of transport have almost the same movement system in common - pedestrians will instinctively react to the current environment, such as avoidance, passing, etc.; the same is true for driverless cars, facing the obstacles in front of the car, they must make judgments t...

AI Technical Summary

Problems solved by technology

Therefore, this shortcoming leads to a limited range of access to the context of the recurrent neural network

In addition, algorithms based on recurrent neural networks and gated recurrent units can only provide the only trajectory prediction path with the greatest possibility based on probability, but they cannot take into account the multimodal possibility of pedestrians’ future trends, which is not In line with the objective law that there may be multiple possible movement trajectories when pedestrians are moving

Therefore, the above methods have not considered comprehensively and exhaustively the problem of pedestrian trajectory prediction in complex scenes.

Images