Posture Estimation from Tactile Signals Using a Masked Forward Diffusion Model

Utilizing tactile sensors embedded in intelligent mats is an attractive non-intrusive approach for human motion analysis. Interpreting tactile pressure 2D maps for accurate posture estimation poses significant challenges, such as dealing with data sparsity, noise interference, and the complexity of mapping pressure signals. Our approach introduces a novel dual-diffusion signal enhancement (DDSE) architecture that leverages tactile pressure measurements from an intelligent pressure mat for precise prediction of 3D body joint positions, using a diffusion model to enhance pressure data quality and a convolutional-transformer neural network architecture for accurate pose estimation. Additionally, we collected the pressure-to-posture inference technology (PPIT) dataset that relates pressure signals organized as a 2D array to Motion Capture data, and our proposed method has been rigorously evaluated on it, demonstrating superior accuracy in comparison to state-of-the-art methods.