A Hybrid Approach of DenseNet121 with Attention and Bi-LSTM for Yoga Pose Estimation
This study presents an Automated Pose Recognition system using Enhanced Chicken Swarm Optimization with Deep Learning (APR-ECSODL), a cutting-edge solution for identifying and categorizing human postures from images and videos with high accuracy. The system is designed to integrate advanced AI techn...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Faculty of Mechanical Engineering in Slavonski Brod, Faculty of Electrical Engineering in Osijek, Faculty of Civil Engineering in Osijek
2025-01-01
|
| Series: | Tehnički Vjesnik |
| Subjects: | |
| Online Access: | https://hrcak.srce.hr/file/477984 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study presents an Automated Pose Recognition system using Enhanced Chicken Swarm Optimization with Deep Learning (APR-ECSODL), a cutting-edge solution for identifying and categorizing human postures from images and videos with high accuracy. The system is designed to integrate advanced AI techniques, providing an innovative approach to pose recognition that leverages several sophisticated machine learning models and algorithms to enhance performance. The pre-processing stage involves applying a Wiener Filter (WF) for effective noise removal, ensuring that the data is clean and ready for analysis. Dynamic Histogram Equalization (DHE) is then employed to enhance image contrast, improving the visibility of key features within the images. For segmentation, the YOLOv8 model is used to isolate relevant regions of interest, providing a precise input for the next phase. Feature extraction is conducted using OpenPose, a widely recognized tool for obtaining key human body points. This step is crucial for capturing detailed information about the postures. The classification of these poses is performed using a Self-Attention Based Gated Recurrent Unit (SA-GRU) model. This model enhances accuracy by incorporating self-attention mechanisms, allowing the system to focus on significant features within the data. Performance optimization is achieved through the Enhanced Chicken Swarm Optimization (ECSO) method, which fine-tunes the parameters of the system to ensure optimal results. The APR-ECSODL technique was rigorously tested on a posture image classification dataset from Kaggle, demonstrating its effectiveness in categorizing various poses. By integrating these cutting-edge deep learning and AI methodologies, the APR-ECSODL system sets a new standard in pose recognition, offering a robust tool for applications in fields such as fitness monitoring, rehabilitation, and human-computer interaction. This approach not only ensures accurate pose identification but also enhances practicing quality and helps prevent errors, making it a valuable asset in diverse domains. |
|---|---|
| ISSN: | 1330-3651 1848-6339 |