Research On Sitting Posture Recognition System Based On Kinect V2

Sitting as the most common posture for adolescents’ daily life and learning has always been widely concerned by scholars.In recent years,the number of people suffering from various diseases caused by poor sitting posture has been increasing,and the age has continued to be younger.Therefore,the study of youngster’s sitting posture recognition and correction can protect their spinal development and avoid a series of diseases such as cervical spondylosis and myopia,so that young people can grow up healthily.With the rapid development of computer technology,depth sensors are also developing faster and faster.Depth sensors are widely used in various fields of research because of their ability to obtain depth information.This project uses the Kinect V2 developed by Microsoft to obtain the joint points of human bones.After the joint point data is processed,different sitting postures are defined and described.Finally,the sitting posture is recognized by the random forest algorithm and the GWO-SVM algorithm,and the recognition is performed at the same time.Prompt and correct the bad sitting posture.Since Kinect has different development platforms based on different systems,and the OpenNI development platform has richer middleware and more mature applications,this topic chooses to develop on the OpenNI development platform.Use OpenNI’s middleware NITE for bone tracking to obtain the required bone joint point information,and calculate the joint rotation angle feature through angle conversion.Aiming at the problem of sitting posture recognition,this paper uses random forest algorithm and GWO-SVM algorithm to recognize sitting posture,and evaluates the performance and recognition rate of the model.Experimental studies have shown that GWO-SVM has a higher recognition accuracy in the face of multi-classification problems with a small number of samples.In order to explore the recognition rate of the system,it is necessary to use a pure sitting data set for training and verification.The existing data set cannot meet this requirement.Therefore,this article builds a pure sitting data set and divides the sitting posture into the correct sitting posture.There are seven kinds of sitting postures and bad postures.At the same time,6 boys of different heights are invited to take the test.Finally,in order to realize the display of the sitting posture and the prompt and correction of the bad sitting posture,this article uses the pyQt5 development framework for development.The application is able to identify and display each detected sitting posture,and at the same time correct and advise the identified bad sitting posture.