Human motion prediction method based on recurrent neural network of adversarial learning

Abstract

The invention relates to a human motion prediction method based on a recurrent neural network of adversarial learning. The method mainly comprises the following steps: firstly, carrying out data preprocessing, and converting data into an Euler angle in a quaternion space to train a network model; and then, constructing a GRU recurrent neural network model. In the model, a training algorithm basedon adversarial learning and a quaternion constraint loss function are adopted to improve the prediction precision of the model, and the stiffness phenomenon generated by a long-time prediction sequence is improved. Tests show that the method can predict the long-time motion trend without causing stiffness of the predicted posture, the predicted motion error is lower than that of other methods, andthe predicted motion trend is more accurate.

Description

technical field [0001] The invention belongs to the fields of computer graphics and human-computer interaction, and in particular relates to a method for human body motion prediction based on adversarial learning-based cyclic neural network. Background technique [0002] In recent years, due to the rapid development of human-computer interaction robots and autonomous driving, the prediction of human future movement trends has attracted more and more attention. Correctly predicting human activities in the future can help robots judge human intentions, assist and respond to human activities, which is very valuable for the development of the field of human-computer interaction. In the field of autonomous driving, correctly judging the intention of pedestrians and taking necessary emergency measures before danger occurs can effectively avoid traffic accidents. [0003] Due to the uncertainty of human future motion and the complex dynamic characteristics, it is still a huge chal...

AI Technical Summary

Problems solved by technology

[0004] Both of these methods can construct human motion behavior, but the traditional mathematical modeling method is more complex and computationally expensive, and is not suitable for large data sets

However, the error accumulation generated when using the cyclic neural network modeling method to predict the long-term sequence makes the predicted sequence easy, and the result of motion prediction becomes unreliable.

Therefore, long-term motion prediction remains one of the biggest challenges in motion prediction

Images