Research On Indoor Human Action Recognition Based On 3D Skeleton

Human action recognition is one of the hot spots of the research in computer vision and pattern recognition field.The goal of human action recognition is to automatically analyze the action of which is being performed from an unknown video or image sequence.At present,human action recognition technology is not only used in monitoring systems,intelligent home,sports analysis,human-computer interaction,film and television production and medical rehabilitation and many other areas,but also created a huge economic and social benefits.In this paper,based on the studying and analyzing the current situation of domestic and foreign research on human motion recognition and related action recognition algorithm,we take the human skeleton model as the research object,this paper studies the human action recognition,we take the human skeleton model as the research object and carry out the research on human action recognition.The main works of this paper are as follows:(1)Based on the analysis of the traditional human body model,we extract the RGBD data through Microsoft Kinect device and study the action of these data representation and recognition methods,extract the human skeleton joints,the establishment of 3D action data set.(2)Aiming at the single-view human action,this paper proposes a method of recognition of human body based on 3D skeleton-based hidden Markov model.Firstly,the extracted 3D skeleton features are vectorized,and then the Baum-Welch algorithm is used to train the parameters of the hidden Markov model and finally identify them(3)A action recognition based on convolution neural network is proposed for single-view human action.The whole process has 3 convolution operations and 2 pooling(down sampling)operations.Each successive 7-frame image passes through these operations to produce a 128-dimensional feature vector that we need.At last,a linear classifier is used to identify.(4)As for multi-perspective situation,we firstly use Kinect device to get videos of human action in front perspective,oblique perspective and side perspective,and extract joint points of bones and get human global features as well as local features of arms and legs which composed 3D bones feature set.And then we take online dictionary learning on the feature set.Finally,we recognize human activity with linear support vector machine LSVM(Linear Support Vector Machine).In this paper,a large number of experimental data are tested and compared with other algorithms.The results show that the algorithm has a good recognition rate.The research work of this paper has good reference significance for the follow-up research work,and has certain application prospects and academic value.