AI-Driven Body Posture Recognition and Predictive Analysis of Long-term Health Risk

This research uses a model-fitting process guided by anatomical structures to estimate human body posture and skeletal system in three-dimensional space using just one RGB (Red Green Blue - refers to color) image. A three-dimensional mesh of a human body is reconstructed by generating the medial axis of an SMPLest-X body model with the help of Voronoi diagrams. The resulting representation of the body provides information about internal support structures without the need for multiple images of the same person from different points of view. When potential medial axis points are determined, a 17 joint, articulated skeleton that resembles the human body's structure is produced. The locations of the joints can then be used to extract key components of the individual's postural position including neck bending, upper body tilting, shoulder symmetry and alignment of the pelvis. In addition to determining these characteristics, this framework can also be used to evaluate whether the individual has a posture that is ergonomically safe (or poses risk for long-term health problems). From a cost-effective and accessible viewpoint, this model has been demonstrated to provide accurate body posture reconstruction and reliable posture-based health assessment, thus providing a viable means for real-time ergonomic monitoring.