基于融合不变性特征与混合核方法的体育视频动作识别  

Sport video motion recognition based on fusing invariant features and hybrid kernel methods

作　　者：刘俊来[1] LIU Jun-lai(Department of Physical Education,Jining University,Jining 273100,China)

机构地区：[1]济宁学院体育系,山东济宁273100

出　　处：《沈阳工业大学学报》2022年第2期198-202,共5页Journal of Shenyang University of Technology

基　　金：山东省科技厅自然科学基金项目(ZR201809370231).

摘　　要：针对体育视频动作识别方法正确率较低的问题,提出了一种结合融合不变性特征与混合核方法的体育视频动作识别方法.采用高斯混合模型构建不变性特征,并对特征进行降维.采用混合核方法分别完成局部特征与全局特征的分类.标准体育动作数据集上的实验结果表明,降维后的融合不变性特征能够保留体育动作关键信息,与混合核方法配合密切,该方法既能够显著提升识别性能,也能够提升识别效率.该方法可以构建实时、在线的体育视频动作识别,且识别效果良好.Aiming at the low accuracy of sport video motion recognition methods,a sport video motion recognition method fusing invariant feature and hybrid kernel was proposed.A mixed Gaussian model was used to construct the invariant features and reduce the feature dimension.The hybrid kernel method was used to classify both local and global features.The experimental results of the standard sport motion data set show that the fusion of invariant feature after dimension reduction can keep the key information of sport motions,which is closely matched with the hybrid kernel method.The as-proposed method can not only improve the recognition performance significantly,but also enhance the recognition efficiency.The as-proposed method can accomplish the real-time and online sport video motion recognition with excellent recognition results.

关 键 词：高斯混合模型 投影降维 核方法 体育视频 动作识别 全局特征 局部特征 支持向量机 

分 类 号：TM343[电气工程—电机]