Design And Implement Of Posture Recognition Software Based On Kinect Sensor

With the development and progress of society,people's work,study and life have undergone tremendous changes.With great pressure,people's work intensity and learning intensity are increasing.Sitting for a long time will be the normal state of work and study.And medicine shows that sitting for long periods of time and poor posture can damage your health.Therefore,effective detection of sitting posture and warning of poor sitting posture will have very practical significance.Currently,posture recognition are mainly composed of single class characteristics,and most papers only algorithm implementation stage,this topic will be starting from the actual application,for sitting position to identify the problem of practical application,implementation can use posture recognition software,based on the device sensor,using mature OpenCV image processing library and convolution neural networks,depth of image acquisition,processing,training and recognition.This software features and advantages: can add any kind of sitting position without considering how to extract features.As a passive sensor,Kinect has little impact on users.The system mainly consists of three modules: image acquisition module,image preprocessing module and sitting posture recognition module.In this project,the three modules are respectively implemented in three programs to reduce the development complexity of each program and improve the debugging efficiency.The three programs are acquisition program,training program and identification program:(1)The acquisition program uses C# language and Microsoft's WPF user interface framework to implement the UI interface,which contains function buttons and a preview window for images.The program obtains the depth image data and body number mask data of the Kinect sensor through Kinect for Windows SDK 2.0.Through the constant exploration of interface writing,interface layout and the research of Kinect driving,the acquisition program was realized,and the staff was called to complete the sample collection of 8 kinds of sitting posture.This provides an important material for training the convolutional neural network.(2)The training program uses OpenCV to realize the function of image preprocessing module,and uses TensorFlow framework to create convolutional neural network.By referring to a large number of existing examples of convolutional neural network,the effects of different network structures were tested and the structure of the network was adjusted.Finally,the training and model output of the neural network were completed.This completes the system and the core algorithm part.(3)The recognition program uses WPF framework to implement the UI interface,uses TensorFlow to load the convolutional neural network model trained,and uses EmguCV to realize the function of image preprocessing module.Finally,the software is implemented to identify 8 major bad sitting postures.