Ultra Inertial Poser: Scalable Motion Capture and Tracking from Sparse Inertial Sensors and Ultra-Wideband Ranging

TL;DR

Ultra Inertial Poser tackles full-body pose estimation from sparse inertial sensors by introducing inter-sensor distances measured with ultra-wideband ranging to limit drift. A graph-based neural model fuses per-sensor 3D states with distance cues to estimate SMPL pose and global translation, trained with AMASS-derived data and evaluated on UIP-DB. The authors demonstrate state-of-the-art accuracy against PIP and TIP, with notable reductions in position error and motion jitter, enabling scalable, wireless motion capture in real-world environments. This approach reduces reliance on camera-based systems and expands motion capture to outdoor or unconstrained settings.

Abstract

While camera-based capture systems remain the gold standard for recording human motion, learning-based tracking systems based on sparse wearable sensors are gaining popularity. Most commonly, they use inertial sensors, whose propensity for drift and jitter have so far limited tracking accuracy. In this paper, we propose Ultra Inertial Poser, a novel 3D full body pose estimation method that constrains drift and jitter in inertial tracking via inter-sensor distances. We estimate these distances across sparse sensor setups using a lightweight embedded tracker that augments inexpensive off-the-shelf 6D inertial measurement units with ultra-wideband radio-based ranging$-$dynamically and without the need for stationary reference anchors. Our method then fuses these inter-sensor distances with the 3D states estimated from each sensor Our graph-based machine learning model processes the 3D states and distances to estimate a person's 3D full body pose and translation. To train our model, we synthesize inertial measurements and distance estimates from the motion capture database AMASS. For evaluation, we contribute a novel motion dataset of 10 participants who performed 25 motion types, captured by 6 wearable IMU+UWB trackers and an optical motion capture system, totaling 200 minutes of synchronized sensor data (UIP-DB). Our extensive experiments show state-of-the-art performance for our method over PIP and TIP, reducing position error from $13.62$ to $10.65cm$ ($22\%$ better) and lowering jitter from $1.56$ to $0.055km/s^3$ (a reduction of $97\%$).

Figures (8)

• Figure 1: Overview of our method. The person is wearing 6 sensor nodes, each featuring an IMU and a UWB radio. Our method processes the raw data from each sensor to estimate sequences of global orientations, global accelerations, and inter-sensor distances. These serve as input into our learning-based pose estimation to predict leaf-joint angles as well as the global root orientation and translation as SMPL parameters.
• Figure 2: Architecture details for the Distance Attention Graph Convolutional Network (DA-GCN). Given node orientation $R_t$ and edge distance $D_t$, DA-GCN estimates sensor positions conforming to the distance measurements. The model consists of a distance attention branch with learnable matrices $A_l$ and $B_l$ to represent the correlation between nodes and a weight-modulation branch $M_l$ to learn disentangled transformation for different nodes.
• Figure 3: Inter-sensor distance RMSE in our dataset.
• Figure 4: Qualitative comparison of pose estimates from the recordings in our dataset. \ref{['fig:compare_long']} shows more examples.
• Figure 5: Qualitative comparisons among different methods on our test set.