Research Of The 3D Grid Human Model Sleep Posture Recognition

For a long time,the 3D model has a huge application space,so it is a very necessary task to obtain the 3D model in the real life.However,the digital 3D model of the real human body reflects the details of the human body's three-dimensional features,which is more difficult to obtain than other objects and is urgent for us.A digital 3D model of real human,is named as computer human,and which is safer,more flexible and convenient,and easy to be storaged,visualized and controlled etc in the computer compared with physical true human body model.And it can be used in all kinds of experimental research of wireless body area network,and the technology development of wireless body area network is not fully mature,which contribute to the development of wireless body area network.The 3D digital visualization of human body model by computer technology in the medical care is applied in the biological mechanics analysis and human motion simulation,to analyze the complexity of the human shoulder joint movement.Now depth camera such as Kinect and so on,which can be quickly and easily to get the 3D information of objects especially the human body,caused a huge wave of research.In the first step,the 3D model of human body with different postures was obtained by scanning using Artec/Kinect et al.The model contains massive cloud point data that is mass feature points in the surface of the human body model,which will lead to more or less repeated feature information between these obtained human 3D point cloud data at the same time,so a simple pretreatment of removing some isolated points,noise points etc is needed.The second step is that the use of visual similarity change the problem of extracting features in the three-dimensional space into the problem of extracting shape feature in two dimensional space,which contain all the three-dimensional geometry features of the model.The light field characteristics in the above process is used to obtain the features information of the 3D digital model,which is robust to model degradation,noise interference and model deformation.Through the establishment of a projected coordinate system,the depth image is acquired from the camera position on the surface located in the large regular polyhedron(the polyhedral space contains the entire 3D human model).Because the image has rich information,we take the corresponding feature extraction algorithm to get the details information.The SIFT algorithm is used to extract the scale features and invariant rotation features of each depth image,and the feature vector contains information such as position,scale,rotation invariant,view change,affine transformation and so on.Through k-means clustering algorithm SIFT features obtained from the previous step was clustered and was encoded into visual words,and visual words generated bag of words,which needs to adopt a good approximation learning model of real problem model to study themselves well.The support vector machine method is established on the learning theory of VC dimension and minimum structural risk principle,and it has the good generalization ability by finding the best compromise between model complexity(i.e.learning accuracy for specific training samples)and learning ability(ability of identifying any samples freely)in the limited sample information.And it has many unique advantages in nonlinear and high dimensional pattern recognition,and can be applied to other machine learning problems such as function fitting.The paper then studied the Pyramid matching histogram intersection kernel support vector machine,and bag of words of test database was identified by the finished learning training model of which,and which provides a little contribution for the study of human-computer interaction,wireless body area network and medical monitoring field.