| Computer vision is about enabling computers to automatically recognize objects and understand the outside world.Human motion recognition based on computer vision has broad application prospects and great economic and social value in the fields of intelligent monitoring,intelligent household appliances,human-machine interface,sports and entertainment motion analysis,content-based video retrieval and other fields.However,how to effectively characterize human motion in video is still a problem in the field of motion recognition.In addition,video motion recognition based on deep learning is also rapidly emerging.This paper uses VGG network in deep convolutional neural network to conduct in-depth research on video human motion recognition.The main research contents of this paper have the following two aspects:(1)A new motion feature extraction algorithm is studied.In order to improve the recognition rate of HOG in motion recognition application,the algorithm calculates the covariance matrix by using the HOG feature as the sample,and then maps the covariance matrix from the Riemannian manifold to the linear space through matrix logarithm operation,and then extracts from the covariance matrix.Descriptors are classified using support vector machines based on different kernel functions.The experimental database is KTH,Weizman and UCF Sports,and the algorithm is implemented on MATLAB2013.(2)A motion recognition method based on deep learning is studied.In this paper,VGG-16 in deep convolutional neural network is used.VGG-16 can make network parameters not too much under the premise of ensuring network depth.Finally,the classifier that classifies human motion features is softmax.In order to improve the quality of the training data set,the video is first processed into a picture frame in MATLAB,and then some of the images that do not contain the target human body are screened out.The implementation platform is python+Caffe.The algorithm get a good recognition rate on the UCF Sports database. |
PDF Full Text Request