Pressure2Motion: Hierarchical Human Motion Reconstruction from Ground Pressure with Text Guidance

TL;DR

Pressure2Motion tackles privacy-preserving motion capture by reconstructing full-body motion from ground pressure and text prompts, eliminating cameras and wearables. It introduces a dual-level pressure feature extractor and a hierarchical diffusion-based synthesizer that injects high-level trajectory and fine-grained posture shifts into a pretrained Motion Diffusion Model, guided by textual semantics. The MPL dataset provides large-scale, text–pressure–motion triplets enabling robust benchmarking, and experiments show state-of-the-art reconstruction quality, physical plausibility, and strong semantic alignment with captions. The work demonstrates a practical, privacy-aware MoCap paradigm with potential for deployment in low-light, in-the-wild environments, while acknowledging limitations in surface diversity and computational demands.

Abstract

We present Pressure2Motion, a novel motion capture algorithm that reconstructs human motion from a ground pressure sequence and text prompt. At inference time, Pressure2Motion requires only a pressure mat, eliminating the need for specialized lighting setups, cameras, or wearable devices, making it suitable for privacy-preserving, low-light, and low-cost motion capture scenarios. Such a task is severely ill-posed due to the indeterminacy of pressure signals with respect to full-body motion. To address this issue, we introduce Pressure2Motion, a generative model that leverages pressure features as input and utilizes a text prompt as a high-level guiding constraint to resolve ambiguities. Specifically, our model adopts a dual-level feature extractor to accurately interpret pressure data, followed by a hierarchical diffusion model that discerns broad-scale movement trajectories and subtle posture adjustments. Both the physical cues gained from the pressure sequence and the semantic guidance derived from descriptive texts are leveraged to guide the motion estimation with precision. To the best of our knowledge, Pressure2Motion is a pioneering work in leveraging both pressure data and linguistic priors for motion reconstruction, and the established MPL benchmark is the first benchmark for this novel motion capture task. Experiments show that our method generates high-fidelity, physically plausible motions, establishing a new state of the art for this task. The codes and benchmarks will be publicly released upon publication.

Figures (10)

• Figure 1: By conditioning on pressure signals and text descriptions, Pressure2Motion reconstructs high-fidelity, physically realistic motions, addressing the challenge of synthesizing human motion from sparse and noisy pressure data.
• Figure 2: Samples from our MPL dataset, each pairing ground pressure and motion with five levels of text descriptions.
• Figure 3: The Pressure2Motion pipeline. We first extract an overall Movement Trajectory $\mathbf{T}_{\text{traj}}$ and fine-grained Posture Shifts $\mathbf{S}_{\text{shift}}$ from pressure maps. These signals are then processed by our adapter branch to provide hierarchical control: a ControlNet encodes the trajectory for global guidance, while an Adapter Block fuses this with posture shifts for local refinement. The resulting features are injected into the pretrained MDM to synthesize plausible motion aligned with the pressure signals.
• Figure 4: Visual comparisons on the MPL dataset. Yellow denotes the predicted results of different methods; blue represents the ground-truth motions. The motions reconstructed by ours align best with the ground truth, especially in the foot region.
• Figure 5: Visualization results of ablation study.