Design And Implementation Of Human Sitting Posture Detection System Based On Machine Vision

Nowadays,the health problems caused by improper sitting posture are increasingly serious.For a specific group of people,maintaining unhealthy posture for a long time not only does harm to the body,but also produces a series of negative psychological problems,which are becoming more and more prominent.Based on this,it is very important to establish sitting posture test and give correct guidance and supervision.Under normal conditions,contact sensors or video monitoring are used to detect the normal sitting posture,but the actual detection process will be affected by a variety of factors.Therefore,on the basis of detection based on machine vision correlation method,combining Ada Boost algorithm and support vector machine algorithm,this paper studies the feature extraction and sitting posture classification,and further studies the recognition of additional actions when partial faces are occluded.In this paper,Ada Boost algorithm and Harr feature are used to train face cascade classifiers and human eye cascade classifiers,and "integral graph" is used to reduce the computation.In addition,the detection of additional facial motion features under occlusion is discussed,and a face feature detection method based on ellipse skin color model to detect chinsupporting condition of hand and Hough line transform to detect pen-containing condition is proposed.Through preprocessing the images collected by the camera in real time,the human eye coordinate features and face area features in the face region are extracted,and the sitting posture feature extraction method and the additional motion feature extraction method in this paper are proposed.Finally,the characteristic data are collected and the missing and abnormal characteristic data are eliminated,which makes the obtained data more effective and reliable.A sitting posture classifier based on support vector machine is designed,and the additional actions are determined.Finally,by changing the parameters of different kernel functions,the best C value of linear kernel is 0.7,the best C value of radial basis kernel is 2.1,and the best G value is 0.001.The average recognition rates of linear kernel function and radial basis function are 99% and 94%,respectively.Therefore,the model of linear kernel function is selected for training and the training results are saved.Using VS2015 and MFC framework,a sitting posture detection system is designed on the computer side,and the above training model is imported.Finally,the recognition rate of sitting posture and additional actions of the tested person is above 94%.Experiments show that the complex background and high or low light intensity have certain influence on the recognition of additional actions,but the recognition rate of sitting posture can still reach more than 94%,which verifies the stability and reliability of the system.To sum up,the system can effectively detect and identify the changes of students' postures,providing guidance for reminding and supervising students to maintain good sitting posture.