3D Human Pose Estimation Using Pressure Images on a Smart Chair

Mingjie Zhao, Fangting Xie, Ziyu Wu, Zhen Liang, Xiaohui Cai

Abstract: Humans spend an amount of time sitting on chairs. Detecting poor sitting pose in time can reduce adverse health effects. Different sitting pose among individuals influence the pressure distribution in the areas where the human body contacts the chair. To explore the potential of utilizing chair pressure data for estimating human pose, we initially collected a dataset called the Temporal Human Sitting Pose (TSP) dataset, comprising 180K synchronized pressure and 3D pose data. Subsequently, we proposed a deep-learning method solely leveraging chair pressure data for 3D human pose estimation. Through experiments validating the method’s feasibility, we achieved a 93.13mm joint position error on our TSP dataset. The results demonstrate that utilizing chair pressure data could be a promising alternative for estimating human pose during sitting.