Research map / 3D vision / Human pose estimation
BodyMAP - Jointly Predicting Body Mesh and 3D Applied Pressure Map for People in Bed
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes BodyMAP, a single network that jointly predicts SMPL body mesh and per-vertex 3D applied pressure for people in bed from depth and pressure images, plus BodyMAP-WS, which learns 3D pressure without 3D pressure labels via a differentiable 2D projection loss.
Problem
Predicting 3D human body pose/shape and the 3D pressure applied on the body for people resting in bed (e.g., under blankets) to help prevent pressure ulcers; prior methods predict only 2D pressure, subsets of body regions, or approximate 3D pressure through multi-stage pipelines.
Approach
A unified ResNet + SMPL + Feature Indexing Module + PointNet network that fuses depth and pressure-image features per mesh vertex to output body mesh and per-vertex pressure times predicted contact; BodyMAP-WS learns 3D pressure implicitly by aligning a differentiable 2D projection of the predicted 3D pressure map with the sensed 2D pressure image.
Key results
- BodyMAP-PointNet with depth and pressure achieves ~25% improvement over BPW on both body mesh and 3D pressure prediction on SLP test subjects.
- BodyMAP-WS, trained without 3D pressure labels, improves 3D pressure error by 8.8% over BPW.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use both depth and pressure images rather than depth alone (ablation) | BodyMAP-PointNet · SLP test subjects 81–102, all blanket conditions | MPJPE (mm): 60.66 ± 1.615 -> 51.01 ± 1.071 | Helped |
| use both depth and pressure images rather than depth alone (ablation) | BodyMAP-PointNet · SLP test subjects 81–102, all blanket conditions | v2vP (kPa2): 2.47 ± 0.034 -> 2.14 ± 0.030 | Helped |
| use both depth and pressure images rather than pressure image only (ablation) | BodyMAP-PointNet · SLP test subjects 81–102, all blanket conditions | MPJPE (mm): 76.54 ± 3.170 -> 51.01 ± 1.071 | Helped |
| use both depth and pressure images rather than pressure image only (ablation) | BodyMAP-PointNet · SLP test subjects 81–102, all blanket conditions | v2vP (kPa2): 2.29 ± 0.014 -> 2.14 ± 0.030 | Helped |
| replace fully connected pressure prediction layers with PointNet (ablation) | BodyMAP · SLP test subjects 81–102, all blanket conditions | v2vP (kPa2): 2.56 ± 0.007 -> 2.14 ± 0.030 | Helped |
| train the BodyMAP-WS pressure model on all poses instead of lateral poses only (ablation) | BodyMAP-WS · SLP supine test poses | v2vP (kPa2): 2.865 -> 2.513 | Helped |
| use BodyMAP-PointNet instead of BPWnet | 3D body mesh prediction · SLP test subjects 81–102, all blanket conditions | MPJPE (mm): 69.36 -> 51.01 ± 1.071 | Helped |
| use BodyMAP-PointNet instead of BPWnet | 3D pressure prediction · SLP test subjects 81–102, all blanket conditions | v2vP (kPa2): 2.84 -> 2.14 ± 0.030 | Helped |
| train BodyMAP-WS pressure prediction on all poses instead of lateral poses only (ablation) | BodyMAP-WS with mesh model pretrained on lateral poses · SLP supine test poses | v2vP (kPa2): 2.865 -> 2.513 | Helped |
Limitations
- Ground truth 3D pressure is approximated by vertical projection, which fails for self-contact
- Pressure on heels and toes is slightly worse for BodyMAP-WS due to contact regularization
Experimental setup
Datasets: BodyPressureSD SLP
Benchmarks: SLP
Metrics: MPJPE PVE anatomical height/chest/waist/hips error v2vP v2vP 1EA v2vP 2EA
Baselines: BPW BPBnet Pyramid Fusion BodyMAP-Conv
Builds on: SMPL PointNet BPW (BodyPressureSD) Pyramid Fusion
Closest papers by meaning
- PI-HMR: Towards Robust In-bed Temporal Human Shape Reconstruction with Contact Pressure Sensing (CVPR 2025 · similarity 0.88)
- DiSRT-In-Bed: Diffusion-Based Sim-to-Real Transfer Framework for In-Bed Human Mesh Recovery (CVPR 2025 · similarity 0.85)
- MotionPRO: Exploring the Role of Pressure in Human MoCap and Beyond (CVPR 2025 · similarity 0.85)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.83)
- MMVP: A Multimodal MoCap Dataset with Vision and Pressure Sensors (CVPR 2024 · similarity 0.82)
- Self-Supervised 3D Human Mesh Recovery from a Single Image with Uncertainty-Aware Learning (AAAI 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.