Rex Tsou

IMP2023 Conference Paper Abstract

Human pose estimation plays a vital role in contemporary technological applications, especially in the domains of motion detection and tracking. However, practical applications often face a significant challenge when it comes to accurately labeling crucial skeletal points on the human body, particularly in cases where these points are obscured or missing. To address these issues, this paper introduces a diffusion model based on generative AI. The primary objective of this model is to accurately predict obscured or missing skeletal points, enhancing the precision of human body pose prediction. The core concept of this model treats the process of predicting human body poses as a denoising procedure, where complete joint point data is considered the accurate representation of the human pose, while missing skeletal points are treated as noise. The generative AI's diffusion model excels at efficiently eliminating this noise, thereby providing more precise results in predicting human body poses. This implies that in practical applications, whether it be in sports analysis, medical imaging, or virtual reality, we can obtain more reliable information regarding human body poses.