Motion data denoising method and system

Abstract

The invention discloses a motion data denoising method and system. The invention provides a motion data denoising method and system. Motion data containing multiple different noises are mixed to serveas input data of the training auto-encoder. The corresponding accurate motion data is used as label data, the joint position error, the bone length error and the smoothness error are used as loss functions to train the autoencoder, and the obtained bidirectional circulation autoencoder can perform real-time denoising on the motion data. According to the denoising method and system provided by the invention, the support of a motion database is not needed in actual use, any type of noise contained in the training data can be removed, and the universality is good. Meanwhile, the self-encoder istrained by taking minimization of joint position errors, minimization of bone length errors and minimization of smoothness errors as optimization targets, and it can be guaranteed that motion data obtained after denoising are free of jitter, natural and real.

Description

technical field [0001] The invention relates to the field of computer three-dimensional animation, in particular to a method and system for denoising motion data. Background technique [0002] Three-dimensional human skeleton motion data is widely used in computer animation, games, virtual reality and motion recognition fields. Most of the motion capture devices currently collecting this type of data often cause the collected data to contain noise or some node data is missing due to various reasons. Miscalibration of equipment positions, sensor noise, and even performers' clothing can also contribute to inaccurate data acquisition. [0003] Traditional hand-edited denoising methods require animators to check motion data frame by frame for noise, which is time-consuming and error-prone. The data-driven motion denoising method is currently the main motion data denoising method, and its main idea is to use the existing noise-free motion database to obtain the correlation betw...

AI Technical Summary

Problems solved by technology

[0004] (1) When denoising, it is necessary to query the noise-free motion in the existing database to denoise the current noise motion, which makes the de-noising system based on this method need to integrate a large-scale motion database during use, making the system occupy storage space. The space is huge, and it is difficult to apply on some devices with small storage scale, such as mobile phones and tablets

[0005] (2) When denoising, only one type of noise can be denoised, which makes the denoising method only applicable to a specific requirement or a specific motion data acquisition device, so this type of method lacks versatility

Images