Sitting Posture Recognition Technology And Application Based On Human Skeleton Point Informatio

In modern society,people face increasing pressure from work and study,often maintaining poor sitting postures for long periods of time,which can easily lead to various health problems.In this context,human body sitting posture recognition has become a popular research field of great concern.This article aims to explore how to accurately identify human body sitting postures and use posture monitoring software to correct poor sitting postures,in order to provide some reference and inspiration for relevant research in the field of human body sitting posture recognition.The main work and innovations include:Firstly,56 volunteers were invited to collect images of 9 types of sitting postures.After screening and discarding some unqualified images,a dataset of 6913 human body sitting posture images was retained.Then,based on the characteristics of different sitting postures,a human body sitting posture skeleton point feature dataset containing 30-dimensional features was constructed based on feature engineering,providing a solid data foundation for subsequent research on sitting posture recognition algorithms.Secondly,a multimodal fusion neural network model was constructed for recognizing human body sitting postures.In this model,two types of network branches were designed to simultaneously process image type data and numerical type data,and the output features of these two branches were concatenated and fused before being sent to the classifier to obtain the final sitting posture recognition result.Experimental results showed that this multimodal fusion model achieved an accuracy rate of 93.85% for recognizing 9 types of sitting postures,which was better than the model that used single type of data input.Thirdly,a human body sitting posture recognition model based on the Alpha Pose human body skeleton point detection algorithm and the ensemble learning stacking strategy was proposed.This model sampled frames at intervals from the video stream and used the Alpha Pose algorithm to quickly process the extracted images to obtain the coordinates of human body skeleton points in the images.Then,based on the characteristics of different sitting postures,30-dimensional feature vectors were constructed using the skeleton point coordinate information,and the Stacking fusion model was used to achieve accurate posture recognition.The test results showed that this model achieved an average recognition rate of 98.55% for 9 types of sitting postures,and had low implementation costs and high application value.Finally,a sitting posture monitoring software based on the Windows platform was designed and implemented.In addition to the sitting posture recognition algorithm,various practical functions such as sitting posture image display,poor sitting posture reminder,custom sitting posture calibration,and further statistical analysis of sitting posture data were added.Through practical application testing,the effectiveness and reliability of this software were verified,which can provide users with more comprehensive sitting posture monitoring services.