{"title":"基于深度学习和可变形部分模型的舞蹈动作特征提取","authors":"Shuang Gao, Xiaowei Wang","doi":"10.4108/eai.5-1-2022.172783","DOIUrl":null,"url":null,"abstract":"In complex scenes, the accuracy of dance movement recognition is not high. Therefore, this paper proposes a deep learning and deformable part model (DPM) for dance movement feature extraction. Firstly, the number of filters in DPM is increased, and the branch and bound algorithm is combined to improve the accuracy. Secondly, deep neural network model is used to sample points of interest according to human dance movements. The features extracted from the DPM and deep neural network are fused. It achieves a large reduction in the number of model parameters and avoids the network being too deep. Finally, dance movement recognition is performed on the input data through the full connection layer. Experimental results show that the proposed method in this paper can get the recognition result more quickly and accurately on the dance movement data set.","PeriodicalId":43034,"journal":{"name":"EAI Endorsed Transactions on Scalable Information Systems","volume":"1 1","pages":"11"},"PeriodicalIF":1.1000,"publicationDate":"2022-01-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Feature extraction of dance movement based on deep learning and deformable part model\",\"authors\":\"Shuang Gao, Xiaowei Wang\",\"doi\":\"10.4108/eai.5-1-2022.172783\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"In complex scenes, the accuracy of dance movement recognition is not high. Therefore, this paper proposes a deep learning and deformable part model (DPM) for dance movement feature extraction. Firstly, the number of filters in DPM is increased, and the branch and bound algorithm is combined to improve the accuracy. Secondly, deep neural network model is used to sample points of interest according to human dance movements. The features extracted from the DPM and deep neural network are fused. It achieves a large reduction in the number of model parameters and avoids the network being too deep. Finally, dance movement recognition is performed on the input data through the full connection layer. Experimental results show that the proposed method in this paper can get the recognition result more quickly and accurately on the dance movement data set.\",\"PeriodicalId\":43034,\"journal\":{\"name\":\"EAI Endorsed Transactions on Scalable Information Systems\",\"volume\":\"1 1\",\"pages\":\"11\"},\"PeriodicalIF\":1.1000,\"publicationDate\":\"2022-01-05\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"EAI Endorsed Transactions on Scalable Information Systems\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.4108/eai.5-1-2022.172783\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q4\",\"JCRName\":\"COMPUTER SCIENCE, INFORMATION SYSTEMS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"EAI Endorsed Transactions on Scalable Information Systems","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.4108/eai.5-1-2022.172783","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q4","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
Feature extraction of dance movement based on deep learning and deformable part model
In complex scenes, the accuracy of dance movement recognition is not high. Therefore, this paper proposes a deep learning and deformable part model (DPM) for dance movement feature extraction. Firstly, the number of filters in DPM is increased, and the branch and bound algorithm is combined to improve the accuracy. Secondly, deep neural network model is used to sample points of interest according to human dance movements. The features extracted from the DPM and deep neural network are fused. It achieves a large reduction in the number of model parameters and avoids the network being too deep. Finally, dance movement recognition is performed on the input data through the full connection layer. Experimental results show that the proposed method in this paper can get the recognition result more quickly and accurately on the dance movement data set.