Related Experiment Video
Updated: Jun 28, 2025
07:25
An Objective and Child-friendly Assessment of Arm Function by Using a 3-D Sensor
Published on: February 12, 2018
6.9K
3D human pose data augmentation using Generative Adversarial Networks for robotic-assisted movement quality
Xuefeng Wang1, Yang Mi2, Xiang Zhang3
1College of Sports, Woosuk University, Jeonju, Republic of Korea.
Frontiers in Neurorobotics
|April 22, 2024
Summary
This study enhances 3D human pose data using Generative Adversarial Networks (GANs) and robot-assisted collection. The approach improves motion recognition accuracy and efficiency for complex human movements.
Area of Science:
- Computer Vision
- Biomechanical Engineering
Background:
- Human motion recognition systems require diverse and complex datasets.
- Current datasets often lack sufficient diversity for rare or complex movements.
Purpose of the Study:
- To introduce an advanced method for augmenting 3D human pose data.
- To improve the quality, diversity, and complexity of motion datasets.
Main Methods:
- Utilized Generative Adversarial Networks (GANs) for synthetic data generation.
- Integrated Support Vector Machine (SVM) for classification and DenseNet for feature extraction.
- Incorporated robot-assisted technology for precise data collection.
Main Results:
- Achieved highly realistic and diverse 3D human motion data generation.
- Demonstrated significant improvements in classification accuracy and data processing efficiency.
- Outperformed traditional methods in motion quality assessment.
Conclusions:
- The integrated network model effectively enhances 3D human pose data augmentation.
- Provides a strong foundation for advancements in human motion recognition.
- Offers practical applications in sports science, rehabilitation, and virtual reality.
