面向动作捕获数据分割的图卷积网络模型

Graph Convolutional Networks for Motion Capture Data Segmentation

提出了一种基于图卷积神经网络的方法来进行动作捕获数据的分割。具体而言,首先引入骨架属性图来表达动作捕获数据序列段中的每一帧;然后建立基于骨架属性图卷积操作的模型来学习动作捕获数据序列段的时空结构。最后利用骨架属性图卷积操作所学到的深度特征通过核化时序切割方法实现对动作捕获数据的分割。实验表明,此方法与其他方法相比是有明显的优势的。

Anovel method is proposed to segment long motion capture data into different actions using graph CNNs.First,an Skeleton Attribute Graph(SAG)construction method is introduced torepresent the spatial relation information for each frame of motion segments.Then,a novel model with SAG graph convolution is constructed to learn spatio-temporal structure of motion segments.Finally,the Kernel-Based Temporal Cut(KBTC)isapplied to segment the given motion capture data based on the deep feature learned by the SAG graph convolution.The experimental results show that the proposed method demonstrates superior performance in comparison to other methods for 3 D motion capture data segmentation.