Quaternion-Based Local Frame Alignment between an Inertial Measurement Unit and a Motion Capture System

Effects of Data Augmentation on the Nine-Axis IMU-Based Orientation Estimation Accuracy of a Recurrent Neural Network

The nine-axis inertial and measurement unit (IMU)-based three-dimensional (3D) orientation estimation is a fundamental part of inertial motion capture. Recently, owing to the successful utilization of deep learning in various applications, orientation estimation neural networks (NNs) trained on large datasets, including nine-axis IMU signals and reference orientation data, have been developed. During the training process, the limited amount of training data is a critical issue in the development of powerful networks. Data augmentation, which increases the amount of training data, is a key approach for addressing the data shortage problem and thus for improving the estimation performance. However, to the best of our knowledge, no studies have been conducted to analyze the effects of data augmentation techniques on estimation performance in orientation estimation networks using IMU sensors. This paper selects three data augmentation techniques for IMU-based orientation estimation NNs, i.e., augmentation by virtual rotation, bias addition, and noise addition (which are hereafter referred to as rotation, bias, and noise, respectively). Then, this paper analyzes the effects of these augmentation techniques on estimation accuracy in recurrent neural networks, for a total of seven combinations (i.e., rotation only, bias only, noise only, rotation and bias, rotation and noise, and rotation and bias and noise). The evaluation results show that, among a total of seven augmentation cases, four cases including 'rotation' (i.e., rotation only, rotation and bias, rotation and noise, and rotation and bias and noise) occupy the top four. Therefore, it may be concluded that the augmentation effect of rotation is overwhelming compared to those of bias and noise. By applying rotation augmentation, the performance of the NN can be significantly improved. The analysis of the effect of the data augmentation techniques presented in this paper may provide insights for developing robust IMU-based orientation estimation networks.

Estimation of human spine orientation with inertial measurement units (IMU) at low sampling rate: How low can we go?

Studying people in their daily life is important for understanding conditions with multi-faceted aetiology such as chronic low back pain. Inertial measurement units can be used to reconstruct the posture and motion of the body outside of laboratories to enable this research. The battery life of these sensors strongly affects the usability of the system, since recharging them frequently is inconvenient and can lead to additional errors. A major determinant of the battery life for these sensors is sampling rate, but the relationship between sampling rate and accuracy in motion reconstruction is not well documented. We measured the spine of 12 participants using inertial measurement units across a variety of tasks such as sitting, standing, walking, and jogging. The orientation of the spine was reconstructed using several filters, including a novel filter developed specifically for high performance at low sampling frequencies. Benchmarking against optical motion capture, we developed a model showing that the error of all tested filters depends exponentially on the sampling frequency, with the optimal filter gains showing a similar exponential relationship. Using this model of error, we developed a criterion for recommending minimum sampling frequencies for accurate motion estimates for each task, finding frequencies ranging from about 13 to 35 Hz sufficient depending on the task. Although we only studied the spine, these models should provide insight into optimizing sampling rate and filter parameters for inertial measurements in general use.

In-vitro validation of inertial-sensor-to-bone alignment

A major shortcoming in kinematic estimation using skin-attached inertial sensors is the alignment of sensor-embedded and segment-embedded coordinate systems. Only a correct alignment results in clinically relevant kinematics. Model-based inertial-sensor-to-bone alignment methods relate inertial sensor measurements with a model of the joint. Therefore, they do not rely on properly executed calibration movements or a correct sensor placement. However, it is unknown how accurate such model-based methods align the sensor axes and the underlying segment-embedded axes, as defined by clinical definitions. Also, validation of the alignment models is challenging, since an optical motion capture ground truth can be prone to disturbances from soft tissue movement, orientation estimation and manual palpation errors. We present an anatomical tibiofemoral ground truth on an unloaded cadaveric measurement set-up that intrinsically overcomes these disturbances. Additionally, we validate existing model-based alignment strategies. Modeling the degrees of freedom leads to the identification of rotation axes. However, there is no reason why these axes would align with the segment-embedded axes. Relative inertial-sensor orientation information and rich arbitrary movements showed to aid in identifying the underlying joint axes. The first dominant sagittal rotation axis aligned sufficiently well with the underlying segment-embedded reference. The estimated axes that relate to secondary kinematics tend to deviate from the underlying segment-embedded axes as much as their expected range of motion around the axes. In order to interpret the secondary kinematics, the alignment model should more closely match the biomechanics of the joint.

Reference in-vitro dataset for inertial-sensor-to-bone alignment applied to the tibiofemoral joint

Skin-attached inertial sensors are increasingly used for kinematic analysis. However, their ability to measure outside-lab can only be exploited after correctly aligning the sensor axes with the underlying anatomical axes. Emerging model-based inertial-sensor-to-bone alignment methods relate inertial measurements with a model of the joint to overcome calibration movements and sensor placement assumptions. It is unclear how good such alignment methods can identify the anatomical axes. Any misalignment results in kinematic cross-talk errors, which makes model validation and the interpretation of the resulting kinematics measurements challenging. This study provides an anatomically correct ground-truth reference dataset from dynamic motions on a cadaver. In contrast with existing references, this enables a true model evaluation that overcomes influences from soft-tissue artifacts, orientation and manual palpation errors. This dataset comprises extensive dynamic movements that are recorded with multimodal measurements including trajectories of optical and virtual (via computed tomography) anatomical markers, reference kinematics, inertial measurements, transformation matrices and visualization tools. The dataset can be used either as a ground-truth reference or to advance research in inertial-sensor-to-bone-alignment.

Evaluation of Inertial Sensor Data by a Comparison with Optical Motion Capture Data of Guitar Strumming Gestures

Computing technologies have opened up a myriad of possibilities for expanding the sonic capabilities of acoustic musical instruments. Musicians nowadays employ a variety of rather inexpensive, wireless sensor-based systems to obtain refined control of interactive musical performances in actual musical situations like live music concerts. It is essential though to clearly understand the capabilities and limitations of such acquisition systems and their potential influence on high-level control of musical processes. In this study, we evaluate one such system composed of an inertial sensor (MetaMotionR) and a hexaphonic nylon guitar for capturing strumming gestures. To characterize this system, we compared it with a high-end commercial motion capture system (Qualisys) typically used in the controlled environments of research laboratories, in two complementary tasks: comparisons of rotational and translational data. For the rotations, we were able to compare our results with those that are found in the literature, obtaining RMSE below 10° for 88% of the curves. The translations were compared in two ways: by double derivation of positional data from the mocap and by double integration of IMU acceleration data. For the task of estimating displacements from acceleration data, we developed a compensative-integration method to deal with the oscillatory character of the strumming, whose approximative results are very dependent on the type of gestures and segmentation; a value of 0.77 was obtained for the average of the normalized covariance coefficients of the displacement magnitudes. Although not in the ideal range, these results point to a clearly acceptable trade-off between the flexibility, portability and low cost of the proposed system when compared to the limited use and cost of the high-end motion capture standard in interactive music setups.

Estimation of 3D Knee Joint Angles during Cycling Using Inertial Sensors: Accuracy of a Novel Sensor-to-Segment Calibration Procedure Based on Pedaling Motion

This paper presents a novel sensor-to-segment calibration procedure for inertial sensor-based knee joint kinematics analysis during cycling. This procedure was designed to be feasible in-field, autonomously, and without any external operator or device. It combines a static standing up posture and a pedaling task. The main goal of this study was to assess the accuracy of the new sensor-to-segment calibration method (denoted as the 'cycling' method) by calculating errors in terms of body-segment orientations and 3D knee joint angles using inertial measurement unit (IMU)-based and optoelectronic-based motion capture. To do so, 14 participants were evaluated during pedaling motion at a workload of 100 W, which enabled comparisons of the cycling method with conventional calibration methods commonly employed in gait analysis. The accuracy of the cycling method was comparable to that of other methods concerning the knee flexion/extension angle, and did not exceed 3.8°. However, the cycling method presented the smallest errors for knee internal/external rotation (6.65 ± 1.94°) and abduction/adduction (5.92 ± 2.85°). This study demonstrated that a calibration method based on the completion of a pedaling task combined with a standing posture significantly improved the accuracy of 3D knee joint angle measurement when applied to cycling analysis.