Motion data redirection method based on neural network

Abstract

The invention discloses a motion data redirection method based on a neural network. The motion data redirection method comprises the following steps of based on motion data represented by joint coordinates, training by taking a motion reconstruction error as a loss function to obtain an auto-encoder; acquiring source motion data, calculating hidden variables corresponding to the source motion datathrough the auto-encoder, and reconstructing motion; applying a bone length constraint, a footprint constraint, a root joint position constraint and a bone angle constraint to the reconstruction motion, and reversely propagating constraint loss to an implicit variable space to optimize the implicit variable; through multiple iterations, the auto-encoder completing motion redirection from the source bone to the target bone. According to the method, the hidden variable corresponding to the source motion data is calculated through the auto-encoder, the motion is reconstructed, then various constraints are applied to the reconstructed motion, motion reconstruction loss is propagated back to the hidden variable space to optimize the hidden variable of the auto-encoder, the redirected motion isfinally obtained through multiple iterations, and a good motion redirection effect is achieved.

Description

technical field [0001] The invention relates to the technical field of motion simulation and animation production, in particular to a neural network-based motion data redirection method. Background technique [0002] Character animation is currently widely used in many applications such as computer animation, games and virtual reality. The animation driving of 3D models is one of the popular research fields in graphics. Among them, the use of bone motion data to drive 3D models is a common method. . Usually, designers design a 3D model and its skeleton model, and then use the existing skeletal motion data to drive the 3D model to form character animation. The methods of obtaining skeletal motion data include manual design and application of motion capture equipment to collect. However, manually designing skeletal motion data is time-consuming and laborious, and obviously cannot be used as a mainstream method. Motion capture data is highly targeted and can only record the ...

AI Technical Summary

Problems solved by technology

Nowadays, more and more convenient devices are used to collect human bone motion data. However, the motion data obtained by these convenient motion collection devices is represented by joint coordinates, and lacks kinematic constraints, making the method based on inverse kinematics unsuitable.

Images