Title: An efficient human action recognition framework based on hybrid features and enhanced long short term memory

Authors: B. Suresh Kumar; S. Viswanadha Raju

Addresses: Computer Science and Engineering, Vardhaman College of Engineering, Hyderabad, 501218, India ' Computer Science and Engineering, JNTU Hyderabad, 501218, India

Abstract: Although the scientific group has given considerable attention, an effective method of detecting human activity in the physical realm remains elusive because of variations in appearance, interactions between objects, and mutual occlusion. So, in this study, an efficient human action recognition (HAR) system based on hybrid features with enhanced long short-term memory (ELSTM) is proposed. Initial key frames are extracted from a sequence of input videos using the structural similarity measure (SSIM). After that, the features namely, coverage factor, space-time interest (STI) points, and shape features are extracted from the key frames. Then, the selected features are fed to the ELSTM classifier to classify a diffident activity of a human. The proposed ELSTM is designed by LSTM with the integration of adaptive golden eagle optimisation (AGEO) in the weight update process to select possible weights. The proposed approach is evaluated in terms of' accuracy, precision, recall, and F-Measure.

Keywords: HAR; human action recognition; AGEO; adaptive golden eagle optimisation; ELSTM; enhanced long short-term memory; STI; space-time interest; coverage factor; shape; SSIM; structural similarity measure.

DOI: 10.1504/IJSSE.2023.133016

International Journal of System of Systems Engineering, 2023 Vol.13 No.3, pp.248 - 270

Received: 29 Aug 2022
Accepted: 10 Oct 2022
Published online: 24 Aug 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article