基于深度神经网络的人体动作识别方法
|
摘要
针对静态图像集中人体动作种类繁杂且识别准确率较低的问题,提出一种基于深度神经网络的人体动作识别方法;该方法采用迁移学习的思想对GoogLeNet模型进行改进,使得网络在预训练之后能够对行为个体的种类具有一定的姿势表达能力;采用逻辑分类中的逻辑回归多分类来实现动作的多分类,并通过建立动作识别模型应用系统对其进行验证;通过MATLAB2017处理平台对该模型进行测试,并得出图像的平均识别率。结果表明,本文中提出的方法在公开的图像数据集PPMI上的平均识别率相对较高,证实了构建的基于GoogLeNet人体动作识别模型应用系统对人体动作的分类是可行且有效的。
In order to solve the problem that the kinds of human action are complex and the recognition accuracy was low in the static image, a method of human action recognition based on the deep neural network was proposed. This method improved the GoogLeNet model by using the idea of migration learning, so that the network had a certain posture express-ion ability for the types of behavioral individuals after the pre-training. The Softmax classification was used to the logical classification to achieve the actions multi-classification, which was validated by an established motion recognition model application system. The new model was tested by MATLAB2017 processing platform, and the average recognition rate of the image was obtained.The results show that the average recognition rate of the new method is relatively high on the open image data set PPMI, which proves that the classification of the human action based on the GoogLeNet human action recognition model is feasible and effective.
In order to solve the problem that the kinds of human action are complex and the recognition accuracy was low in the static image, a method of human action recognition based on the deep neural network was proposed. This method improved the GoogLeNet model by using the idea of migration learning, so that the network had a certain posture express-ion ability for the types of behavioral individuals after the pre-training. The Softmax classification was used to the logical classification to achieve the actions multi-classification, which was validated by an established motion recognition model application system. The new model was tested by MATLAB2017 processing platform, and the average recognition rate of the image was obtained.The results show that the average recognition rate of the new method is relatively high on the open image data set PPMI, which proves that the classification of the human action based on the GoogLeNet human action recognition model is feasible and effective.
引文
[1] 李瑞峰,王亮亮,王珂.人体动作行为识别研究综述[J].模式识别与人工智能,2014,27(1):35-48.
[2] 王鑫,沃波海,管秋,等.基于流形学习的人体动作识别[J].中国图象图形学报,2014,19(6):914-923.
[3] 程乐,周抒,宋艳红,等.应用于图像分割的改进贪婪蛇算法[J].济南大学学报(自然科学版),2018,32(3):212-217.
[4] 刘吉庆.基于视频的人体动作识别方法研究[D].济南:山东大学,2013.
[5] 沃波海.基于深度数据的人体动作识别[D].杭州:浙江工业大学,2014.
[6] 石祥滨,刘拴朋,张德园.基于关键帧的人体动作识别方法[J].系统仿真学报,2015,27(10):2401-2408.
[7] 耿驰.基于深度学习的人体动作识别[D].南京:南京邮电大学,2016.
[8] 张建国.基于深度学习的场景分类[D].锦州:辽宁工业大学,2016.
[9] AZARY S,SAVAKIS A.A spatiotemporal descriptor based on radial distances and 3D joint tracking for action classification[C]//IEEE International Conference on Image Processing.Orlando:IEEE,2013:769-772.
[10] DELAITRE V,LAPTEV I,SIVIC J.Recognizing human actions in still images:a study of bag-of-features and part-based representations[C]//British Machine Vision Conference,BMVC 2010.Aberystwyth:Proceedings DBLP,2010:1-11.
[11] SIMONYAN K,ZISSERMAN A.Two-stream convolutional networks for action recognition in videos[J].Computational Linguistics,2014,1(4):568-576.
[12] 韩敏捷.基于深度学习的动作识别方法研究[D].南京:南京理工大学,2017.
[13] MAJI S,BOURDEV L,MALIK J.Action recognition from a distributed representation of pose and appearance[J].IEEE Computer Society Conference on Computer Vision & Pattern Recognition,2012,32(14):3177-3184.
[14] THURAU C,HLAVAC V.Pose primitive based human action recognition in videos or still images[C]//IEEE Conference on Computer Vision and Pattern Recognition,2008.Anchorage:IEEE,2008:1-8.
[15] WANG Y,JIANG H,DREW M S,et al.Unsupervised Discovery of Action Classes[C]//IEEE Conference on Computer Vision and Pattern Recognition,2006.New York:IEEE,2006:1654-1661.
[16] SZEGEDY C,LIU W,JIA Y,et al.Going deeper with convolutions[C]//IEEE Conference on Computer Vistion and Pattern Recognition,2015.Boston:IEEE,2015:1-9.
[17] 白阳,万洪林,白成杰.基于GoogLeNet的静态图像中人体行为分类研究[J].电脑知识与技术,2017,13(18):186-188.
[18] ESCORCIA V,NIEBLES J C,GHANEM B.On the relationship between visual attributes and convolution networks[C]//IEEE Conference on Computer Vision and Pattern Recognition.Boston:IEEE,2015:1256-1264.
[19] 钱学成.静态图像中的行为识别系统设计[D].广州:华南理工大学,2016.
[20] 彭沛沛.基于集合表示的图像分类[D].哈尔滨:哈尔滨工程大学,2016.
[21] KIM T E,KIM M H.Improving the search accuracy of the VLAD through weighted aggregation of local descriptors[J].Journal of Visual Communication and Image Representation,2015,31:237-252.
[22] WANG X,LU H.Action recognition with uncertain VLAD[C]//Seventh International Symposium on Computational Intelligence and Design.Hangzhou:IEEE,2015:185-188.
[23] LIU Z,WANG S,TIAN Q.Fine-residual VLAD for image retrieval[J].Neurocomputing,2016,173:1183-1191.
[2] 王鑫,沃波海,管秋,等.基于流形学习的人体动作识别[J].中国图象图形学报,2014,19(6):914-923.
[3] 程乐,周抒,宋艳红,等.应用于图像分割的改进贪婪蛇算法[J].济南大学学报(自然科学版),2018,32(3):212-217.
[4] 刘吉庆.基于视频的人体动作识别方法研究[D].济南:山东大学,2013.
[5] 沃波海.基于深度数据的人体动作识别[D].杭州:浙江工业大学,2014.
[6] 石祥滨,刘拴朋,张德园.基于关键帧的人体动作识别方法[J].系统仿真学报,2015,27(10):2401-2408.
[7] 耿驰.基于深度学习的人体动作识别[D].南京:南京邮电大学,2016.
[8] 张建国.基于深度学习的场景分类[D].锦州:辽宁工业大学,2016.
[9] AZARY S,SAVAKIS A.A spatiotemporal descriptor based on radial distances and 3D joint tracking for action classification[C]//IEEE International Conference on Image Processing.Orlando:IEEE,2013:769-772.
[10] DELAITRE V,LAPTEV I,SIVIC J.Recognizing human actions in still images:a study of bag-of-features and part-based representations[C]//British Machine Vision Conference,BMVC 2010.Aberystwyth:Proceedings DBLP,2010:1-11.
[11] SIMONYAN K,ZISSERMAN A.Two-stream convolutional networks for action recognition in videos[J].Computational Linguistics,2014,1(4):568-576.
[12] 韩敏捷.基于深度学习的动作识别方法研究[D].南京:南京理工大学,2017.
[13] MAJI S,BOURDEV L,MALIK J.Action recognition from a distributed representation of pose and appearance[J].IEEE Computer Society Conference on Computer Vision & Pattern Recognition,2012,32(14):3177-3184.
[14] THURAU C,HLAVAC V.Pose primitive based human action recognition in videos or still images[C]//IEEE Conference on Computer Vision and Pattern Recognition,2008.Anchorage:IEEE,2008:1-8.
[15] WANG Y,JIANG H,DREW M S,et al.Unsupervised Discovery of Action Classes[C]//IEEE Conference on Computer Vision and Pattern Recognition,2006.New York:IEEE,2006:1654-1661.
[16] SZEGEDY C,LIU W,JIA Y,et al.Going deeper with convolutions[C]//IEEE Conference on Computer Vistion and Pattern Recognition,2015.Boston:IEEE,2015:1-9.
[17] 白阳,万洪林,白成杰.基于GoogLeNet的静态图像中人体行为分类研究[J].电脑知识与技术,2017,13(18):186-188.
[18] ESCORCIA V,NIEBLES J C,GHANEM B.On the relationship between visual attributes and convolution networks[C]//IEEE Conference on Computer Vision and Pattern Recognition.Boston:IEEE,2015:1256-1264.
[19] 钱学成.静态图像中的行为识别系统设计[D].广州:华南理工大学,2016.
[20] 彭沛沛.基于集合表示的图像分类[D].哈尔滨:哈尔滨工程大学,2016.
[21] KIM T E,KIM M H.Improving the search accuracy of the VLAD through weighted aggregation of local descriptors[J].Journal of Visual Communication and Image Representation,2015,31:237-252.
[22] WANG X,LU H.Action recognition with uncertain VLAD[C]//Seventh International Symposium on Computational Intelligence and Design.Hangzhou:IEEE,2015:185-188.
[23] LIU Z,WANG S,TIAN Q.Fine-residual VLAD for image retrieval[J].Neurocomputing,2016,173:1183-1191.