摘要
针对舞蹈视频与动作识别技术相结合的问题,文中研究探讨一种基于计算机视觉的舞蹈视频动作识别技术。该技术首先将获取到的舞蹈视频图像进行灰度化、背景消除和滤波去噪等预处理操作后,得到该视频序列中人物的动作特征。然后利用支持向量机SVM从对象特征样本集中抽取一部分数据样本用于模型的学习训练,训练完成后再对其他部分进行动作分类识别。KTH动作数据库与实拍舞蹈视频的仿真测试结果说明,该方法能够迅速、有效地识别出舞蹈视频中所出现的动作,且平均识别准确率在85%以上,验证了该技术应用于舞蹈动作识别中的可行性。
Aiming at the combination of dance video and motion recognition technology,this paper studies and discusses a dance video motion recognition technology based on computer vision.Firstly,after graying,background elimination and filtering denoising of the obtained dance video image,the motion features of the characters in the video sequence are obtained.Then,a part of the data samples are extracted from the feature samples set of the object by support vector machine(SVM)for model learning and training.After training,the other parts are classified and recognized.The simulation results of KTH action database and real dance video show that the method can recognize the actions in dance video quickly and effectively,and the average recognition accuracy is more than 85%.The feasibility of applying this technology to dance action recognition is verified.
作者
毕雪超
BI Xue chao(Xi’an Vocational and Technical College of Aeronautics and Astronautics,Xi’an 710089,China)
出处
《电子设计工程》
2020年第7期151-155,共5页
Electronic Design Engineering
基金
西航职院2018年度科研计划项目(18XHGZ-011)。
