Research On Human Posture Recognition Based On Convolutional Neural Network

Human posture recognition is one of the most challenging research fields in computer vision field.It is widely used in human-computer interaction,intelligent monitoring,virtual reality,human behavior analysis and other fields.With the introduction of Kinect depth sensing equipment by Microsoft Corp,more and more scholars have paid more attention to the attitude recognition technology based on depth image.However,in the study of the image by the light,shadow and posture change outside the restricted category,difficulty of feature selection,machine learning and recognition algorithm performance limitations and other factors,research progress of human gesture recognition method of gesture recognition is slow,become a research hotspot in recent years.At present,human gesture recognition method used mostly skeleton information and depth image information based on Kinect,based on the image segmentation to achieve human foreground extraction,using HOG,HIFT or the skeleton key coordinates on body posture feature extraction and dimension reduction,finally the model for training and classification.Although these methods result in the recognition accuracy and the accuracy is better,but the body because of high complexity,through the feature extraction of skeletal key points and depth image information often dimension is relatively high,the feature extraction of depth image mostly before also need to carry through image preprocessing,feature extraction and recognition in difficult efficiency high.Therefore,a method of human posture recognition based on convolutional neural network is proposed in this paper.The main research contents and tasks are as follows:(1)outlines the imaging principle of depth image and obtaining method,through the Kinect shooting persons of different size in different backgrounds,illumination images,construct the training sample contains ten positions and test samples,to ensure the diversity of database.(2)the related theories of artificial neural networks and convolutional neural networks are introduced.Introduced the structure and BP neural networks of the classical algorithm,the convolutional neural network structure of a typical case,describes the composition and training mode of network structure and convolutional neural network related knowledge such as activation function,loss function,gradient calculation,softmax regression classification technology.(3)according to the convolutional neural network with local perceptions,weight sharing and network model of low complexity,the image can be directly used as the input of the network and to translation,zoom,tilt deformation network structure has the advantages of high degree of invariance,this paper studies the human gesture recognition based on convolutional neural network,and from the network layer,the number of filters and activation function of the optimization of network structure,classification and recognition to human body posture.