Estimating Lower Limb Kinematics Using a Reduced Wearable Sensor Count

Estimating Lower Limb Kinematics using a Lie Group Constrained Extended Kalman Filter with a Reduced Wearable IMU Count and Distance Measurements

“…The proposed algorithm, L7S (for Lie seven segment), estimates the orientation of the pelvis, thighs, shanks, and feet (i.e., 7 segments) with respect the world frame, W , using either two or three IMUs. It extends the model and assumptions from our prior work [9,29] (L5S-3I, CKF-3I, that aim to estimate the kinematics of five body segments, and places IMUs on the pelvis and shanks). Two variants of the algorithm are described: L7S-3I which uses three IMUs attached at the sacrum and feet (Fig.…”

“…Such configurations can improve user comfort while also reducing setup time and system cost. However, utilizing fewer sensors inherently reduces the amount of kinematic information available; this information must be inferred by enforcing mechanical joint constraints [9], making dynamic balance assumptions, or using additional sensors (e.g., cameras or distance measurement [10,11]). Amongst additional sensor approaches, video-inertial systems are the most common, where IMU measurements help resolve orientation ambiguity for OMC systems [12][13][14][15].…”

“…Recent literature has shown that sparse motion capture is also possible in 3D (e.g., window-based optimization on full body segments linked by 24 ball and socket joints [24]). In our recent work, we tracked five body segments (i.e., the pelvis, thigh, and shanks) using IMUs at the pelvis and ankles using a constrained Kalman filter (CKF) where orientation was represented using quaternions [9]. Building on prior work on state estimation using a Lie group representation ( [25][26][27] for propagating uncertainty, [28] for IMC systems under OSPS configuration), we further extended their work by representing and tracking pose using Lie groups, specifically the special Euclidean group, SE(3) [29].…”

Magnetometer Robust Deep Human Pose Regression With Uncertainty Prediction Using Sparse Body Worn Magnetic Inertial Measurement Units

“…Andrews et al [34] have used an inverse dynamic solver for joint torques and internal/contact forces which satisfies motion priors and sparse sensor measurements, and thus generates physically plausible human motion. The data-driven approaches using reduced sensors (≤ 6 instead of 13-17) [10][11][12][13] are well suited for online implementation than optimization or constrained stochastic filtering [15][16][17][18][19]. But good scalability of data driven approaches typically require a large dataset.…”

“…In past research [7][8][9], a small set of inertial sensors is shown to estimate 3D pose to a reasonable accuracy. The data-driven approaches using reduced sensors (≤ 6 instead of [13][14][15][16][17] [10][11][12][13] are more suitable for ambulatory data capture than full kinematic approach [14][15][16][17][18][19]. The scalability of data driven approach to 3D human pose estimation using reduced sensor set has been demonstrated using deep learning [20] and a large synthetic dataset.…”