RECOGNITION OF HUMAN ACTIONS USING MACHINE LEARNING METHODS

Abstract

Human action recognition is a significant area of focus within computer vision, and it is intertwined with several other disciplines such as computer science, psychology, and healthcare. This is due to the increasing number of videos and the potential applications for automatic video analysis, such as video surveillance, human-machine interaction, sports analysis, and video search. In this research, we applied machine learning algorithms such as Random Forest, MLP Classifier, AdaBoost, and QDA to recognize human actions and compared the results. The results of the tests showed that the MLP Classifier had an accuracy of 97%, the Random Forest had an accuracy of 95%, the AdaBoost had an accuracy of 76%, and the QDA had an accuracy of 74%. In the training dataset, the MLP Classifier had an accuracy of 98%, the Random Forest had an accuracy of 99%, the AdaBoost had an accuracy of 76%, and the QDA had an accuracy of 74%. Out of all the algorithms, the MLP Classifier showed the best results.