Human action classification is a significant issue in the computer vision field. To retrieve essential information from a large number of videos, understanding the content of the videos is very important. We propose an approach that classifies human actions based on the coordinate information of the body parts. The extracted key coordinate points from each frame based on real-time pose estimation algorithms are stored in a csv file. This csv file is inputted into our various files. In this paper, we investigate 4 different approaches, namely: angle heuristics, logistic regression, random forest classifier and deep learning/neural networks to classify the human poses into 3 classes: standing, sitting, lying. Based on our experiments, we are able to deduce that the neural networks perform the best, followed by the random forest classifier, logistic regression and finally the angle heuristics approach. Hence, we recommend these models to be used in future human activity classification systems like fall detection systems, bed-exit systems, etc as they have achieved relatively high accuracy and are generally quite effective. Hence with these models we developed, it would help to improve efficiency of current systems which involve human activity classification using pose estimation.
Done at A*STAR I2R