AUTHOR=Wang Xuefeng , Mi Yang , Zhang Xiang 

TITLE=3D human pose data augmentation using Generative Adversarial Networks for robotic-assisted movement quality assessment

JOURNAL=Frontiers in Neurorobotics

VOLUME=Volume 18 - 2024

YEAR=2024

URL=https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2024.1371385

DOI=10.3389/fnbot.2024.1371385

ISSN=1662-5218

ABSTRACT=3D human pose data augmentation is a technique to enrich and enhance the quality of original datasets through synthetic data, crucial for improving the performance of human motion recognition systems. However, current research still faces shortcomings in terms of diversity and complexity of data, particularly when dealing with rare or complex human movements.Generative Adversarial Networks play an essential role in this field, capable of generating highly realistic and diverse data to overcome these limitations. In our research, we introduce an innovative GANs-SVM-DenseNet network model, coupled with robot-assisted technology to enhance the precision and efficiency of data collection. Our model utilizes GANs to generate realistic 3D human motion data, Support Vector Machine (SVM) for effective classification, and DenseNet for key feature extraction. This integrated approach not only elevates the performance of the data augmentation process but also enhances the model's capability to handle complex data. Experimental results demonstrate that our model exhibits exceptional performance in motion quality assessment, particularly in processing and analyzing complex human movements.Compared to traditional methods, our model shows significant advancements in classification accuracy and data processing efficiency. These findings validate the efficacy of our model, laying a solid foundation for future applications and research in related fields. This study not only provides new perspectives and methods for 3D human orientation data enhancement techniques, as well as robust technical support for practical applications in fields like sports science, rehabilitation medicine, and virtual reality. Our work showcases how the combination of advanced algorithms and robotic technology can effectively address key challenges in data augmentation and motion quality assessment, paving new avenues for future research and development in these domains.