Research On Sitting Posture Recognition Based On Random Forest Algorithm

Good sitting posture is of great significance to people's physical and mental health.Nowadays,due to the needs of work and study,people spend more and more time sitting every day.Sitting for a long time makes it difficult for people to maintain a good sitting posture all the time.The longer the time to maintain a bad sitting posture,the more likely it will lead to a series of health problems,such as myopia,cervical spondylosis and lumbar spondylosis.In this paper,a random forest algorithm based on particle swarm optimization is proposed,and it is applied to the problem of human sitting posture recognition to improve the recognition accuracy,which is conducive to the sitting posture correction system,which can better distinguish the sitting posture and give full play to its functions such as reminder correction.This paper analyzes the morphological characteristics of people's daily sitting posture,and uses random forest classifier to classify and recognize seven common sitting posture,such as good sitting posture,head left deviation,head right deviation,body left deviation,body right deviation,left cheek supporting and right cheek supporting.Firstly,this paper compares and analyzes three common moving target detection algorithms,and finally uses KNN background modeling based background subtraction method to complete the sitting target extraction.Then,the basic principles of PHOG feature and LBP feature are introduced,and these two features are fused serially to improve the representation ability of sitting target.Then,the basic principle of random forest is introduced in detail,including the basic theory of decision tree,the idea of ensemble learning and the construction process of random forest.Aiming at the problem of parameter selection of random forest,an improved particle swarm optimization algorithm is proposed,which mainly improves the inertia weight?and learning factorc₁?c₂ in the iterative formula of particle swarm optimization,in order to make the random forest more stable the number of decision trees T,the depth of decision trees D and the size of attribute feature subset N are more suitable.Finally,this paper uses the random forest algorithm based on particle swarm optimization to train and recognize the sitting posture,and finally obtains an average recognition rate of 98.75%,and then compares it with the random forest algorithm based on support vector machine,decision tree and standard particle swarm optimization to classify and recognize the sitting posture data,which proves the superiority of this algorithm.In this paper,PHOG features and LBP features are fused in series,and combined with the improved particle swarm optimization random forest classification algorithm,the classification and recognition of sitting posture is realized.The final experimental results have achieved good classification effect,which broadens the idea and provides a method for the later research of contactless sitting posture recognition.This method has strong engineering practicality.