LTRACN: A Method for Single Human Activity Recognition
Prashant Sharma, Amartya Mishra, Nilutpol Kashyap, Muheed Muzamil, Rahul Singh Rawat, Ali Imam Abidi, Lokendra Singh Umrao · 2023
In recent years, we have seen that two-stream models (dealing with both the spatial and temporal domain) have performed well and also have achieved state-of-the-art performance with time-series data specifically when working with videos for activity recognition, activity prediction, video captioning, etc. In this paper, we propose a single human activity recognition architecture, Long-Tenn Recurrent Attention Convolutional Network (LTRACN), based on Convolutional Neural Network (CNN), Attention Model, and Long Short-Term Memory (LSTM). The CNN is used to extract the spatial features from the RGB video frames. An Attention model is added to learn which parts in a particular frame are important for the prediction of classes and later on the information is being processed by the LSTM and prediction is done.