Performance of pre-trained model in human activity recognition from videos
Rahul Kumar · 2025
Human Activity Recognition (HAR) problems are earning the attention of researchers due to their wide areas of application, like monitoring, healthcare, sports action analysis, and abnormal activity detection. HAR problems have a large amount of data, which requires lots of computation power. In this study, we use recent transfer learning (TL) models to evaluate performance in the HAR area. These models require less data for training and are easier to employ than traditional deep learning (DL) models. These model performances were evaluated based on precision, recall, F1-score, and accuracy.