Dynamic Human Activity Recognition with Vision-Based Pose Estimation and Machine Learning for Various Age Groups

Human activity recognition is evident in medical institutions, survey systems, security monitoring, and human-computer interaction study fields, This article is related to classifying six basic human-centered behaviors (Punch, stand, sitting, squat, wave, and kick) by implementing machine learning algorithms, Logistic Regression CV and Random Forest algorithm, and SVM classifier Machine learning algorithms are used for the human activity classification. We have used a vision-based algorithm using skeleton pose estimation for feature extraction. We have presented results across age groups of adolescence, adult, and middle age. The dataset used is obtained from a public repository and handled using Open Pose (Kaggle). Human activity computing is an up-and-coming research to interpret individual behavior and make an effort to understand its societal context. The results show the mapping of face and body coordinates on the raw camera feed and produce a probabilistic output and a single output defining one of the six activities (Punch, squat, kick, sit, stand, and wave) on the screen. The result is shown in a screen feed of a fixed rate pixel and works simultaneously as the work runs as an infinite loop till the work is closed hence the results are dynamic and keep changing as per real-time scenarios precise, demanding, and agreeable application of collecting context by detecting human body motion information and thereby making intelligent decisions on real-time data is implemented. Here, activity recognition databases are considered publicly available as repositories. We get an on-screen result based on probability which is displayed on the video frame screening where six actions and three age groups (children, young adults. elderly persons) are displayed.