基于特征动作序列的动态手势识别方法  

Dynamic Gesture Recognition Method Based on Feature Motion Sequence

作　　者：吴佳[1] 刘道星 唐文妍 刘宁[1] 李坤[1] WU Jia;LIU Dao-xing;TANG Wen-yan;LIU Ning;LI Kun(College of Automation and Electronic Information,Xiangtan University,Xiangtan 411105,China)

机构地区：[1]湘潭大学自动化与电子信息学院,湖南湘潭411105

出　　处：《控制工程》2021年第10期1977-1982,共6页Control Engineering of China

摘　　要：针对数据噪声与手势书写不规范性易导致手势识别错误的问题,提出了基于特征动作序列的动态手势识别方法。首先,通过预处理将原始数据转换为两种变换特征作为特征动作识别的依据。其次,对变换特征使用多中心模糊C均值聚类(MCFCM)算法,自适应、无监督地提取特征动作,并将特征动作及对应聚类中心与手势特征动作序列的编码信息保存到知识库。最后,采用改进编辑距离(proED)计算待测特征动作序列与知识库中手势编码信息的相似度,通过相似度匹配实现手势在线识别。将所提方法应用于数字手势识别,识别精度达到98%,实验结果验证了所提方法的有效性。Aiming at the problem that data noise and non-normative gesture writing may easily lead to gesture recognition errors,a dynamic gesture recognition method based on feature motion sequence is proposed.Firstly,the raw data are preprocessed and converted to two features as the basis for the feature motion recognition.Secondly,the multi-center fuzzy C-means cluster(MCFCM)algorithm is used to extract the feature motion adaptively and unsupervisedly,and then the clustering centers,feature motion and encode information of gesture feature motion sequences are saved into the knowledge base.Finally,an improved editing distance(proED)is used to calculate the similarity of feature motion between the test sequences and the encoding information in the knowledge base.Online gesture recognition is realized by similarity matching.The proposed method is applied in digital gesture recognition and obtains 98%recognition accuracy,which shows the effectiveness of the proposed method.

关 键 词：手势识别 特征动作 多中心模糊C均值聚类 改进编辑距离 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]