A One-Shot Learning Method for Human Activity Recognition Using Extracted Essential Poses from Push-Fixed Adaptive Resonance Theory

The utilization of Human Activity Recognition (HAR) through surveillance cameras has the potential to enhance the daily monitoring of elderly individuals living alone. However, existing research in this area predominantly relies on deep learning methods that necessitate large amounts of data for effective training. The acquisition of new data on older adults can be both challenging and costly. To address these limitations, this paper proposes a one-shot learning HAR method that extracts human poses from video data. The study introduces three key techniques: firstly, the PF-ART algorithm is employed to identify the essential poses for each activity. Secondly, a scoring system is designed to provide similarity scores for input activities. Finally, two decision rules are established to distinguish between upper and lower activities. The proposed system is evaluated using the KTH dataset, and the obtained results demonstrate its superior performance compared to previous methods.