Modified Long Short Term Memory Technique for Human Action Recognition from Videos
Boddupally Janaiah, Suresh Pabboju · 2024
Of late, computer vision such as human action recognition from videos became an important research area. This research has assumed significance due to the emergence of artificial intelligence (AI). Since it is a computer vision application, deep learning models are widely used for human action recognition from videos. However, existing methods such as Convolutional Neural Network (CNN) and Long Short Term Memory (LSTM) have problem in dealing with spatio-temporal data. Combining convolution layers and LSTM cell has potential to improve representational power. Towards this end, we proposed a deep learning framework which exploits our proposed hybrid model known as ConvLSTM to leverage human action recognition performance. We proposed an algorithm named Hybrid Deep Learning for Human Action Recognition (HDL-HAR) which exploits ConvLSTM model. We used UCF50 dataset for empirical study. We evaluated performance of ConvLSTM model with existing deep learning models such as CNN and LSTM. Experimental results revealed that the proposed model outperforms the state of the art with highest accuracy 95.23%.