WheelPoser: Sparse-IMU Based Body Pose Estimation for Wheelchair Users

TL;DR

WheelPoser tackles the challenge of upper-body pose estimation for wheelchair users with a real-time, four-sensor sparse-IMU system. It combines a learning-based kinematics module trained with synthetic data (AMASS) and fine-tuned on the WheelPoser-IMU dataset with a physics-based optimizer to produce smooth, torque-aware poses, achieving $14.30^{\circ}$ angular error and $6.74\ \text{cm}$ position error—over 3x improvements over prior sparse-IMU methods. The approach is validated on 167 minutes of wheelchair-related motion and demonstrated in live 60 Hz demos, plus a range of applications from training to inclusive gaming and care. This work advances practical, privacy-preserving motion capture for wheelchair users and points to future directions including more unobtrusive sensing, richer physics, and large-scale inclusive datasets.

Abstract

Despite researchers having extensively studied various ways to track body pose on-the-go, most prior work does not take into account wheelchair users, leading to poor tracking performance. Wheelchair users could greatly benefit from this pose information to prevent injuries, monitor their health, identify environmental accessibility barriers, and interact with gaming and VR experiences. In this work, we present WheelPoser, a real-time pose estimation system specifically designed for wheelchair users. Our system uses only four strategically placed IMUs on the user's body and wheelchair, making it far more practical than prior systems using cameras and dense IMU arrays. WheelPoser is able to track a wheelchair user's pose with a mean joint angle error of 14.30 degrees and a mean joint position error of 6.74 cm, more than three times better than similar systems using sparse IMUs. To train our system, we collect a novel WheelPoser-IMU dataset, consisting of 167 minutes of paired IMU sensor and motion capture data of people in wheelchairs, including wheelchair-specific motions such as propulsion and pressure relief. Finally, we explore the potential application space enabled by our system and discuss future opportunities. Open-source code, models, and dataset can be found here: https://github.com/axle-lab/WheelPoser.

Figures (11)

• Figure 1: WheelPoser uses four IMUs strategically placed on the user's forearms and head, as well as their wheelchair.
• Figure 2: Sample orientation measurements (Euler angles) of two IMUs attached to the user's pelvis and the wheelchair's central axle. The orientations are highly correlated, affirming the close relationship between the pelvis and wheelchair axle motion.
• Figure 3: Correlation coefficient matrices displaying normalized accelerations of the left wrist (LW), right wrist (RW), head (H), pelvis (P), and wheelchair (C), where, for instance, LW-C represents the normalized acceleration of the left wrist with respect to the wheelchair. normalized accelerations with respect to the pelvis and wheelchair are highly correlated.
• Figure 4: Overview of the physics-based optimization module: It calculates reference joint angular acceleration from the kinematics module's estimates. These accelerations are then refined with physical constraints and double-integrated to produce optimized poses.
• Figure 5: The WheelPoser-IMU data collection setup includes four IMUs placed on both the participant and their wheelchair, along with 60 retroreflective markers attached across bony landmarks on the participant's body.