Dynamic Human Action Recognition and Classification Using Computer Vision
Year: 2023
Type: article
Abstract: Action Recognition is one of the time series classification problem where we analyse the data from series of time steps for classifying and predicting the action being performed accurately. Video analysis shows significant improvement as it can predict the outcome of the future state by inferring the present state. This has been possible with the help of Computer Vision, Machine Learning, and Deep Learning fields. In video-based action recognition tasks, many actions can be identified and inferred based on the movements of the actions performed. In our paper, we use UCF50 dataset which consists of fifty categories of actions performed from YouTube. In our paper, we use the video classification model to solve action recognition by analysing each frame of the videos. We aim to create a 2D Convolutional Neural Network classifier, then implement a single frame video classifier along with moving average technique on the live videos, to classify and predict the action performed. Our proposed model has obtained an accuracy of 98.56%.
Authors: Geeta C Mara, Syed Thouheed Ahmed, Vinaya R M
Institution: REVA University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7220164480