摘要
针对传统方法在人体行为识别方面拓展性不强等问题,提出一种序列化的研究思想,提取骨骼图的特征矢量,用SVM训练和识别静态动作,形成序列即可表示动态动作,因此只要丰富静态动作库,就可以实现多种动态动作的识别,具有很好的拓展性。为了减少静态动作识别错误产生的影响,提出一种基于前后信息的纠错算法。实验表明,该算法具有较高的识别准确率,并且具有很好的鲁棒性和实时性。
In view of the fact that the traditional method is not expanding well in human behavior recognition,this paper proposes a serialization research idea.A sequence which can represent the dynamic action is generated by using SVM to train and recognize static action whose feature vectors of skeleton map extracts from Kinect.Therefore,as long as the static action library is rich,a variety of dynamic actions can be identified,and it has good scalability.In order to reduce the influence of the error recognition of static motion,this paper proposes an error correction algorithm based on front and back information.Experiments show that the algorithm has higher recognition accuracy,and has better robustness and real-time.
作者
胡青松
张亮
Hu Qingsong;Zhang Liang(Internet of Things Perception Mine Research Center,China University of Mining and Technology,Xuzhou 221008,China;School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221008,China)
出处
《电子技术应用》
2018年第4期122-125,129,共5页
Application of Electronic Technique
基金
江苏省自然科学基金(BK20151148)
国家重点研发计划资助项目(2017YFC0804409)
关键词
动作识别
骨骼图
特征矢量
SVM
action recognition
skeleton map
feature vectors
SVM
