基于人体行为识别的VR环境下动作分割算法  

Motion Segmentation Algorithm in VR Environment Based on Multi⁃dimensional State Cloud Model

作　　者：金淼 张军 王昊[2] 黄天富 郭志伟 陈习文 王春光 卢冰 周玮 陈卓 王旭 JIN Miao;ZHANG Jun;WANG Hao;HUANG Tianfu;GUO Zhiwei;CHEN Xiwen;WANG Chunguang;LU Bing;ZHOU Wei;CHEN Zhuo;WANG Xu(China Electric Power Research Institute,Wuhan 430074,China;Institute of Electromagnetics,Chinese Academy of Metrology,Beijing 100013,China;Marketing Service Center of State Grid Fujian Electric Power Co.,Ltd.,Fuzhou 350013,China)

机构地区：[1]中国电力科学研究院有限公司,武汉430074 [2]中国计量科学研究院电磁所,北京100013 [3]国网福建省电力有限公司营销服务中心,福州350013

出　　处：《高压电器》2022年第3期163-170,共8页High Voltage Apparatus

基　　金：国家电网公司科技项目(基于人体行为识别的计量现场作业安全管控关键技术研究)。

摘　　要：在虚拟现实环节中的培训类、操作类的系统中,对于动作的标准性、规范性评价已经成为一个日益突出的问题。将培训师进行培训过程的标准动作分割是非常重要和关键的一环,分割的精确与否直接影响后续的培训评价。文中提出了一种基于多维状态云模型预测的方法,首先采用卷积神经网络对操作动作的多传感器数据集进行模型训练预测得到下一时刻的动作多个结果形成多结果数据集;然后对预测的多结果数据集设计多维云模型;最后通过实际动作在云模型中的分布得到对应概率进行动作分割点判定。结果表明基于卷积神经网络的多维云模型的动作分割方法能够比较好地选取分割点,提高动作的划分精度,能够用于虚拟现实环境中的连续动作分割。In the system of training and operation in virtual reality,the standard and normative evaluation of action has become an increasingly prominent problem.It is very important and critical to segment the standard actions of the trainer in the training process.The accuracy of segmentation directly affects the subsequent training evaluation.In this paper,a kind of method based on multi⁃dimensional state cloud model prediction is proposed.Firstly,convolu⁃tional neural network is used to model the multi⁃sensor data set of operation actions,and the multi⁃result data set of the next moment is formed by training and predicting the multi⁃sensor data set of operation actions;Then,a multi⁃di⁃mensional cloud model for the predicted multi⁃result data set is designed;Finally,the actual actions are used in the cloud model.The corresponding probability of the distribution in the pattern is used to determine the action segmenta⁃tion point.The results show that the adaptive multi⁃dimensional cloud model based on convolutional neural network(CNN)can select the segmentation points better,improve the accuracy of action segmentation,and can be used for continuous action segmentation in virtual reality environment.

关 键 词：虚拟现实 培训 动作分割 多维云模型 卷积神经网络 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术] TP391.41[自动化与计算机技术—计算机科学与技术]