Human Action Recognition based on Hybrid Deep Learning Model and Shearlet Transform

• Published in: 2020 12th International Conference on Information Technology and Electrical Engineering (ICITEE) pp. 152 - 155
• Main Author: Al-Azzawi, Nemir Ahmed
• Format: Conference Proceeding
• Language: English
• Published: IEEE 06-10-2020
• Subjects: Classification Accuracy; Deep learning; Feature extraction; Gated Recurrent Unit; Human Action; Image segmentation; Logic gates; Recurrent neural networks; Segmentation; Shearlet Transform; Training; Transforms
• DOI: 10.1109/ICITEE49829.2020.9271687

The hybrid deep learning model has become common in all recent studies dealing with machine vision and human action recognition. Most of the accuracy in revealing knowledge of machine vision is in extracting important features, including segmentation of the image. This paper proposes a new model for recognizing human actions from video sequences by integrating repetitive, gated recurrent neural networks across multiple scales with shearlet-based image segmentation extraction. Segmentations are the most critical information to distinguish human action. The feature extraction can impact the complexity of the calculation and the performance of the algorithm. The idea is to increase training robustness and improve segmentation through the use of the shearlet transform. Hence, the video classification based on a recurrent neural network and shearlet transform will work optimally. The proposed approach is evaluated on human activity videos using KTH, UCF-101, and UCF Sports Action datasets. The experimental results showed state-of-the-art performance in comparison to current methods. The average resulting classification accuracy is 95.1% for the KTH datasets. That was the optimal case in our proposed model reached.