LSTM Method for Human Activity Recognition of Video Using PSO Algorithm

A most alluring method is Human Activity Recognition (HAR) from video, cost-effective computer vision-based recognition research and it has wide range of applications such as Closed-Circuit Television (CCTV) surveillances, gaming and health care unit. The recognition of human activity labeled by the appearance and motions observed from the video. The rise of social informatization, the prominent information technology mentioned by machine vision is applied to many more scenes. The conventional methods of HAR facing problems such as slower recognition and precision rate are low. With the advancement of deep learning approach, the difficulties faced in HAR are addressed. For this analysis, various kinds of neural network have been recommended by researchers and they did HAR effectively. This study also proposes a neural network so called Long Term Short Memory (LSTM) combined with Particle Swarm Optimization (PSO) to predict human activities more accurately in a video. The dataset chosen for this study includes variety of actions including such as pushups, mixing, kayaking, playing basketball, drumming, etc. This approach improves the temporal and spatial dimensions to maximize the recognition rate. The accuracy obtained by processing UCF-50 dataset is relatively higher than the existing approach.