3D human motion prediction method based on depth state space model

Abstract

The invention discloses a 3D human body motion prediction method based on a depth state space model, and the method comprises the steps: taking the position and speed of human body motion as observation, and extracting a motion dynamics rule of a historical posture sequence through a depth network to initialize the state of the state space model; and recursively predicting a plurality of future poses of 3D human motion through state-to-observation transitions. By utilizing the advantages of the deep network and the state space model, the human body motion system is modeled into the deep statespace model, a unified description is provided for various human body motion systems, and meanwhile, the existing model can be analyzed.

Description

Technical field[0001]The present invention belongs to the field of human motion prediction, and more particularly to a 3D human motion prediction method based on a depth state spatial model.Background technique[0002]Humans' understanding and interaction of real world depends on the ability to predict the surrounding environment. Similarly, interactive interactive smart robots must have the ability to predict human future dynamics, so that robots can quickly respond to human changes. Most prediction models generally use only L2Or MPJPE (Average Each Point Position Error) Optimizes the prediction model for all frames or all joints of future postures. Since the loss value in the early time is less than the loss value in the late time, these models are implicitly focused on the prediction of late time, and ignore this hidden relationship between early times and later times, that is, in the recursive model, Early predictions were easily forecast for later moments. Therefore, these models...

AI Technical Summary

Problems solved by technology

Since the loss value at the early moment is smaller than the loss value at the later moment, these models implicitly focus on the prediction of the later moment, while ignoring this hidden relationship between the early moment and the later moment, that is, in the recursive model, Early predictions tend to affect predictions at later moments

Therefore, none of these models can achieve more accurate predictions, especially in the recursive prediction model

Images