基于多特征组合的动态手势识别  

Dynamic gesture recognition based on multi-feature combination

作　　者：曹海婷 戎海龙 焦竹青[1] 马正华[1] CAO Hai-ting1 , RONG Hai-long2, JIAO Zhu-qing1, MA Zheng-hua1(1. School of Information Science and Engineering, Changzhou University, Changzhou 213164, China; 2. School of Urban Rail Transit, Changzhou University, Changzhou 213164, Chin)

机构地区：[1]常州大学信息科学与工程学院,江苏常州213164 [2]常州大学城市轨道交通学院,江苏常州213164

出　　处：《计算机工程与设计》2018年第6期1727-1732,共6页Computer Engineering and Design

基　　金：江苏省自然科学基金项目(BK20140265)

摘　　要：为提高动态手势的识别效果,提出一种基于多特征组合的动态手势分类方法。对表面肌电(surface electromyography,SEMG)和加速度计(accelerometer,ACC)传感器进行特征水平上的融合,分别对两类传感器提取多种类型特征并组合,通过实验对比分析选出最优特征组合。为对短时间肌肉收缩有较好连续性,采用样本熵对表面肌电检测活动段起始点,以隐马尔可夫模型(hidden Markov model,HMM)对手势动作进行识别。实验结果表明,采用最优特征组合后,5名受试者对10类动态手势获得(94.13±1.07)%的平均识别率,有效提高了手势分类准确性。To improve the recognition effects of dynamic gestures,a method of dynamic gestures classification based on multi-feature combination was proposed.Surface electromyography（SEMG）and accelerometer（ACC）sensors were fused at the feature level.Many types of features were extracted and combined from two types of sensors,and through experimental analysis,the best feature combination was determined.To achieve a good continuity of short-term muscle contraction,the sample entropy was proposed to detect activity segments within SEMG signals.The hand gestures were identified using hidden Markov model（HMM）.Experimental results show that the average recognition rate of 10 types of dynamic gestures is up to（94.13±1.07）%over 5 subjects by using the optimal feature combination.The proposed method can effectively improve the classification accuracy of gestures.

关 键 词：表面肌电 加速度计 特征组合 隐马尔可夫模型 样本熵 手势识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]