面向肌电信号的虚拟现实提线木偶动画研究  

Research on Puppet Animation Controlled by Electromyography(EMG) in Virtual Reality Environment

作　　者：谭宇彤 周旭峰 孔令芝 王醒策[1] 武仲科[1] 税午阳[1] 付艳[1] 周明全[1] Vladimir KORKHOV Luciano Paschoal GASPARY TAN Yu-Tong;ZHOU Xu-Feng;KONG Ling-Zhi;WANG Xing-Ce;WU Zhong-Ke;SHUI Wu-Yang;FU Yan;ZHOU Ming-Quan(College of Artificial Intelligence, Beijing Normal University, Beijing 100875, China)

机构地区：[1]北京师范大学人工智能学院,北京100875 [2]Department of Computer Modeling and Multiprocessor Systems, St.Petersburg State University (SPBU) 199034, Russia [3]Institute of Informatics, Federal University of Rio Grande do Sul (UFRGS) 15064, Brazil

出　　处：《软件学报》2019年第10期2964-2985,共22页Journal of Software

基　　金：国家重点研发计划政府间重点专项金砖国家合作项目(2017YFE0100500);国家科技支撑计划(2017YFB1002604,2017YFB1402100; 2017YFB 1002804);北京市自然科学基金(4172033)~~

摘　　要：泉州提线木偶属于首批中国非物质文化遗产,是中华传统文化的实现形式之一.然而,由于木偶体积庞大,携带与操作不便,直接限制了其受众群体.为了实现提线木偶的有效传承与保护,设计了基于手势识别的虚拟现实提线木偶动画方案,构建了基于 MYO 臂环肌电信号的人体生理信号控制动画原型系统,应用两个用户实验验证了算法的高精确性与易操控性.首先,通过低通滤波与平滑实现多通道肌电信号数据的信号处理.其次,提取八通道时域特征与时频域特征,并通过线性判别器将其降维为六维特征向量,实现特征间关联性消除与算法鲁棒性增强.最后,构造多分类支撑向量机实现特征向量,确定手势识别结果.实验验证算法离线动作平均识别准确率为 95.59%,实时动作平均识别准确率达到 90.75%,在 1.1s 左右完成手势识别.面向提线木偶任务,构造了两个用户体验任务,普通用户人群中,木偶动作识别率较高,用户使用意愿、易学性等方面,系统性能亦显著高于真实木偶操控;专业用户在承认系统可用性的同时,具有较高的接受度.用户任务表明该设计满足了手势识别实时性和准确性的要求,具有良好的交互性和趣味性.相关研究可以广泛地应用于计算机动画等类似的系统,对于体验和保护提线木偶具有现实意义.Quanzhou puppet is one of the intangible cultural heritages of China. It is the physical embodiment of traditional Chinese culture. However, the large size of the puppet and inconvenience to carry and manipulate directly makes it hard to reach a wider audience. In order to realize the effective inheritance and protection of Quanzhou puppet, this study designs a virtual real-line puppet animation scheme based on gesture recognition, builds a prototype system which uses MYO Armband EMG signal to control the generation of animation, and applies it in user experiment to verify the high accuracy and easy manipulation of the algorithm. Firstly, low-pass filtering and smoothing is used to process the original multi-channel EMG data. Secondly, after eight-channel EMG signal time-domain feature and time-frequency-domain feature extraction, the dimension of the feature vector is reduced to six by linear discriminator to eliminate the correlation between features and enhance the robustness of the algorithm. Thirdly, a multi-class support vector machine is constructed which uses feature vector to determine the result of gesture recognition. Experiments show that the average recognition accuracy of offline action is 95.59%, the average recognition accuracy of real-time action is 90.75%, and the gesture recognition is completed within 1.1 s. For the puppet task, two users task is designed: the common users and the expert users. In the common user study, the gestures recognition accuracy is high. In the aspects of user’s willingness to use and easiness to learn, the performance of this system is significantly higher than real puppets manipulation. In the expert user study, user’s acceptance and usability of the system are also highly evaluated. These two user tasks indicate the system meets the requirements of real-time and accuracy, and has good interactivity and interesting. Relevant research can be widely applied to similar systems, such as computer animation. It has practical significance for experiencing and protecting

关 键 词：MYO 臂环 EMG 信号处理 手势识别 SVM LDA 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]