Search

Monocular 3D human pose estimation technologies have the potential to greatly increase the availability of human movement data. The best-performing models for single-image 2D-3D lifting use graph convolutional networks (GCNs) that typically require some manual input to define the relationships between different body joints. We propose a novel transformer-based approach that uses the more generalised self-attention mechanism to learn these relationships within a sequence of tokens representing joints. We find that the use of intermediate supervision, as well as residual connections between the stacked encoders benefits performance. We also suggest that using error prediction as part of a multi-task learning framework improves performance by allowing the network to compensate for its confidence level. We perform extensive ablation studies to show that each of our contributions increases performance. Furthermore, we show that our approach outperforms the recent state of the art for single-frame 3D human pose estimation by a large margin. Our code and trained models are made publicly available on Github., Comment: Accepted to ICPR 2022

Computer vision/deep learning-based 3D human pose estimation methods aim to localize human joints from images and videos. Pose representation is normally limited to 3D joint positional/translational degrees of freedom (3DOFs), however, a further three rotational DOFs (6DOFs) are required for many potential biomechanical applications. Positional DOFs are insufficient to analytically solve for joint rotational DOFs in a 3D human skeletal model. Therefore, we propose a temporal inverse kinematics (IK) optimization technique to infer joint orientations throughout a biomechanically informed, and subject-specific kinematic chain. For this, we prescribe link directions from a position-based 3D pose estimate. Sequential least squares quadratic programming is used to solve a minimization problem that involves both frame-based pose terms, and a temporal term. The solution space is constrained using joint DOFs, and ranges of motion (ROMs). We generate 3D pose motion sequences to assess the IK approach both for general accuracy, and accuracy in boundary cases. Our temporal algorithm achieves 6DOF pose estimates with low Mean Per Joint Angular Separation (MPJAS) errors (3.7{\deg}/joint overall, & 1.6{\deg}/joint for lower limbs). With frame-by-frame IK we obtain low errors in the case of bent elbows and knees, however, motion sequences with phases of extended/straight limbs results in ambiguity in twist angle. With temporal IK, we reduce ambiguity for these poses, resulting in lower average errors., Comment: Project page: https://kevgildea.github.io/KinePose/

The mechanical properties of human soft tissue are crucial for impact biomechanics, rehabilitation engineering and surgical simulation. Validation of these constitutive models using human data remains challenging and often requires the use of non-invasive imaging and inverse finite element (FE) analysis. Post processing data from imaging methods such as tagged magnetic resonance imaging (MRI) can be challenging. Digital Image Correlation (DIC) however is a relatively straightforward imaging method and thus the goal of this study was to assess the use of DIC in combination with FE modelling to determine the bulk material properties of human soft tissue. Indentation experiments were performed on a silicone gel soft tissue phantom. A two camera DIC setup was then used to record the 3D surface deformation. The experiment was then simulated using a FE model.

This study presents and validates a novel (non-ECG-triggered) MRI sequence based on SPAtial Modulation of the Magnetization (SPAMM) to non-invasively measure 3D (quasi-static) soft tissue deformations using only six acquisitions (three static and three indentations). In current SPAMM tagged MRI approaches data is typically constructed from many repeated motion cycles. This has so far restricted its application to the measurement of highly repeatable and periodic movements (e.g. cardiac deformation). In biomechanical applications where soft tissue deformation is artificially induced, often by indentation, significant repeatability constraints exist and, for clinical applications, discomfort and health issues generally preclude a large number of repetitions.

A SPAMM tagged MRI methodology is presented allowing continuous (3.3-3.6 Hz) sampling of 3D dynamic soft tissue deformation using non-segmented 3D acquisitions. The 3D deformation is reconstructed by the combination of 3 mutually orthogonal tagging directions, thus requiring only 3 repeated motion cycles. In addition a fully automatic post-processing framework is presented employing Gabor scale-space and filter-bank analysis for tag extrema segmentation and triangulated surface fitting aided by Gabor filter bank derived surface normals. Deformation is derived following tracking of tag surface triplet triangle intersections. The dynamic deformation measurements were validated using indentation tests (~20 mm deep at 12 mm/s) on a silicone gel soft tissue phantom containing contrasting markers which provide a reference measure of deformation. In addition, the techniques were evaluated in-vivo for dynamic skeletal muscle tissue deformation measurement during indentation of the biceps region of the upper arm in a volunteer., Comment: Medical Physics 2012

This paper discusses tension-compression asymmetry properties of Ogden hyperelastic formulations. It is shown that if all negative or all positive Ogden coefficients are used, tension-compression asymmetry occurs the degree of which cannot be separately controlled from the degree of non-linearity. A simple hybrid form is therefore proposed providing separate control over the tension-compression asymmetry. It is demonstrated how this form relates to a newly introduced generalised strain tensor class which encompasses both the tension-compression asymmetric Seth-Hill strain class and the tension-compression symmetric Ba\v{z}ant strain class. If the control parameter is set to q=0.5 a tension-compression symmetric form involving Ba\v{z}ant strains is obtained with the property {\Psi}({\lambda}_1,{\lambda}_2,{\lambda}_3 )={\Psi}(1/{\lambda}_1 ,1/{\lambda}_2 ,1/{\lambda}_3 ). The symmetric form may be desirable for the definition of ground matrix contributions in soft tissue modelling allowing all deviation from the symmetry to stem solely from fibrous reinforcement. Such an application is also presented demonstrating the use of the proposed formulation in the modelling of the non-linear elastic and transversely isotropic behaviour of skeletal muscle tissue in compression (the model implementation and fitting procedure have been made freely available). The presented hyperelastic formulations may aid researchers in independently controlling the degree of tension-compression asymmetry from the degree of non-linearity, and in the case of anisotropic materials may assist in determining the role played by, either the ground matrix, or the fibrous reinforcing structures, in generating asymmetry., Comment: 20 pages