High physical reality sense human body motion capturing method based on neural motion control

Abstract

The invention provides a high-physical-reality-sense human body motion capturing method based on neural motion control. According to the method, firstly, a sampling distribution prior network based on a physical engine is provided, and accurate sampling distribution prior is obtained through training; secondly, providing a scene contact constraint, inputting a single-view video, and obtaining a human body reference motion through an optimization frame; and finally, estimating sampling distribution from the human body reference motion and the current state of the physical role by using the trained sampling distribution prior, and further realizing human body motion capture with high physical reality sense in a non-differentiable physical engine by using a sampling control method. According to the human body capture frame based on neural motion control, hard physical constraint is provided by using a physical engine, unreal physical phenomena such as mold penetrating and shaking in traditional human body motion capture are avoided, and the human body capture frame is convenient to collect, low in cost and easy to implement.

Description

technical field [0001] The invention belongs to the field of computer vision and computer graphics; in particular, it relates to a high physical reality human motion capture method based on neural motion control. Background technique [0002] Human motion capture has a variety of applications in human-computer interaction, personal health management, and human behavior understanding. With the rapid development of markerless motion capture, the market requirements for human motion capture continue to increase. A large body of existing work can capture accurate human pose kinematically from single-view videos and images through network regression or optimization methods. However, due to visual ambiguity, single-view motion capture based on kinematics will cause a series of unnatural phenomena, such as die-through, shaking, and unreasonable sliding of footsteps. Some methods add physical constraints to the reconstruction framework through simple and differentiable physical mo...

AI Technical Summary

Problems solved by technology

However, such methods have large approximation errors; other methods use non-differentiable physics engines and deep reinforcement learning methods for estimation

But training an ideal policy requires a complex configuration process, and the trained policy is very sensitive to environmental changes

In addition, traditional sampling-based control methods require accurate reference motion, generating a reasonable sampling distribution is time-consuming and suffers from random errors

Images