Elsevier

Microelectronics Journal

Volume 45, Issue 12, December 2014, Pages 1603-1611
Microelectronics Journal

Reliable orientation estimation for mobile motion capturing in medical rehabilitation sessions based on inertial measurement units

https://doi.org/10.1016/j.mejo.2014.05.018Get rights and content

Abstract

Fully mobile and wireless motion capturing is a mandatory requirement for undisturbed and non-reactive analysis of human movements. Inertial sensor platforms are used in applications like training session analysis in sports or rehabilitation, and allow non-restricted motion capturing. The computation of the required reliable orientation estimation based on the inertial sensor RAW data is a demanding computational task. Therefore, an analysis of the computational costs and achievable accuracy of a Kalman filter and a complementary filter algorithm is provided. Highly customized and thus low-power, wearable computation platforms require low-level, platform independent communication protocols and connectivity. State-of-the-art small sized commercial inertial sensors either lack the availability of an open, platform independent protocol, wireless connectivity or extension interfaces for additional sensors. Therefore, an extensible, wireless inertial sensor called Institute of Microelectronic Systems Inertial Measurement Unit (IM)2SU, featuring onboard inertial sensor fusion, for use in home based stroke rehabilitation is presented. Furthermore, a Quaternion based, singularity free orientation estimation accuracy error measure is proposed and applied. To evaluate orientation estimation accuracy an optical system is used as golden reference. Orientation estimation based on a Kalman filter and a complementary filter algorithm is evaluated. The proposed IMU provides high orientation estimation accuracy, is platform independent, offers wireless connection and extensibility and is low cost.

Introduction

Motion capturing is used in a wide range of applications, ranging from rehabilitation to virtual reality [1], [2]. Highly reliable camera-based systems require a complex stationary setup and a permanent line of sight between tracked objects and multiple cameras. In contrast, mobile and wearable inertial sensors can be easily attached to the human body [2]. This is mandatory for human motion capturing in sport training sessions or medical applications. They also overcome the spatial limitations of a stationary camera setup.

The target application of the proposed wireless inertial sensor platform is motion capturing during rehabilitation sessions, where the captured movement data is used to generate an audio feedback [3]. Recent research showed that patients in stroke rehabilitation remarkably benefit from this so called sonification of movements [4]. Rehabilitation training tasks include drinking from a glass of water. The established inertial sensor system comprises a wireless sensor network of up to 10 inertial sensors fixed to the patient׳s body to capture complex upper body movements and force sensors at the fingers to detect grasping activities. Orientation estimation based on the data of the IMUs attached to the upper arm and forearm allows the computation of upper limb positions, angles between limbs and velocities using forward kinematics and a connected rigid chain body model.

The system architecture shown in Fig. 1 incorporates sensors attached to the upper arm and forearm, the wearable computation platform and wireless audio feedback via headphones or hearing aids. Additional force sensors for the assessment of the performed grasping tasks could be integrated via expansion interfaces in future.

The software architecture shown in Fig. 1 incorporates a graphical user interface. IMU data processing tasks are data acquisition, sonification and sound synthesis. For sound synthesis the Sound Synthesis Toolkit (STK) [5] framework providing a variety of signal generators is used.

Available sensor platforms are highly integrated and hence small-sized and lightweight [6], [7]. As discussed in Section 2, state-of-the-art commercial and academic inertial sensor platforms are limited regarding extensibility, platform independence, wireless connectivity or accuracy, respectively.

Therefore, a wireless, low power sensor platform using a platform independent protocol is presented in this paper. The following aspects highlight the innovation of the approach:

  • -

    low power RF module and 8-bit MCU.

  • -

    full access via platform independent interfaces.

  • -

    real-time onboard orientation estimation.

In human motion capturing, common algorithmic approaches for orientation estimation like integration, vector observation, complementary filtering, or Kalman filtering are applied. These algorithms work on IMU RAW data acquired by tri-axial gyroscopes, accelerometers and magnetometers. Orientation estimation accuracy is evaluated based on Euler angles, thus suffering from singularities. Therefore, a Quaternion based measure is proposed and additionally provided within the orientation estimation accuracy evaluation.

Furthermore, the following topics regarding orientation estimation are presented.

  • -

    Quaternion based error measure for reliable orientation estimation accuracy assessment.

  • -

    Reduced complexity Kalman filter.

  • -

    Comparison of reduced complexity Kalman filter and complementary filter orientation estimation accuracy.

The paper is organized as follows:

Section 2 presents related work, highlighting previously developed wireless IMUs in academia and industry. The hardware and software architecture of the proposed (IM)2SU are presented in Section 3. An introduction to Kalman filter and complementary filter based IMU orientation estimation and a comparison of the computational costs of two exemplarily implementations are shown in Section 4. Section 5 presents the Euler angle and the proposed Quaternion error measure for orientation estimation accuracy evaluation. The results of the system and sensor fusion latency and orientation estimation accuracy evaluation of the (IM)2SU sensor system are shown in Section 6. Conclusions are drawn in Section 7.

Section snippets

Related work

From the application point of view, sonification on mobile platforms based on inertial sensors is explored in [8], [9]. A mobile system for improving running mechanics is developed in [8]. The system comprises a mobile phone and tri-axial accelerometers and gyroscopes connected via Bluetooth. The sensor is worn at the sacrum and accelerometer data is captured. The computed runner׳s average center of mass is provided as an objective running technique feedback. Due to limited computing

Mobile, wireless inertial measurement unit

The system architecture of the Institute of Microelectronic Systems Inertial Measurement Unit (IM)²SU is split into a hardware part and a firmware part. In the next two sections the hardware aspects and firmware implementation are presented.

Reliable orientation estimation based on IMU data

Besides sensor fusion techniques for orientation estimation based on inertial sensor data, like gyroscope integration, and vector observation, Kalman filters are optimal estimators with respect to the minimization of the error covariance [17]. In the literature, there are various proposals for Kalman filters used in inertial sensor fusion [18]. The filters basically differ in the state vector size and preprocessing steps. An alternative approach to Kalman filter based sensor fusion in the

Error measure for IMU orientation estimation accuracy

A common way of analyzing the orientation estimation accuracy of IMUs is to use an optical tracking system as golden reference [19], [20]. Therefore, the first step is to attach a sufficient number of markers to the sensor in order to achieve reliable tracking by the optical system. The number of markers depends on the optical system configuration, e.g. the number of cameras and the 3D measurement volume.

The wooden measurement piece used for tracking the Xsens MTx and the proposed (IM)²SU is

Evaluation of the proposed IMU

This evaluation section considers sensor fusion algorithm latency and latency induced by the communication protocol applied. Furthermore, the orientation estimation accuracy is evaluated based on a data-set comprising consecutive movements along roll, pitch, and yaw axis sampled at 100 Hz. Fig. 12 shows the captured movement.

The sensor is rotated within a range of approximately±50° along each axis with significant time in resting pose between each rotation. Furthermore, in the time period

Discussion and conclusions

According to the bandwidth limitation the achievable sampling rate for up to four (IM)²SU units using onboard sensor fusion is ≈50 Hz. Compared to the Xsens MTx sensor the proposed (IM)²SU achieves a slightly increased orientation estimation accuracy, while providing full access to the sensor data at a significant lower price. Additionally, the proposed IMU allows the connection of further force sensors required for grasping task assessment. In general, an orientation estimation accuracy of 2.1°

Acknowledgments

The authors thank Rochus Nowosielski and Henning Kluge for their contributions in design and initial operation of the proposed IMU. Furthermore, the authors thank Prof. Rosenhahn and Florian Bauman from the Institut für Informationsverarbeitung at Leibniz Universität Hannover for support and making the Vicon optical motion tracking system available.

References (25)

  • E. R. Bachmann, R. B. McGhee, X. Yun, M. J. Zyda, Inertial and magnetic posture tracking for inserting humans into...
  • H. Zheng et al.

    Position-sensing technologies for movement analysis in stroke rehabilitation

    Medical and biological engineering and computing

    (2005)
  • H.-P. Brückner, M. Wielage, H. Blume, Intuitive and interactive movement sonification on a heterogeneous RISC/DSP...
  • I. Wallis, T. Ingalls, T. Rikakis, L. Olson, Y. Chen, W. Xu, H. Sundaram, Real-time sonification of movement for an...
  • G. Scavone, P. Cook, RtMidi, RtAudio, and a synthesis toolkit (STK) update, in: Proceedings of the International...
  • Xsens Technologies BV,...
  • Shimmer Research,...
  • M. Eriksson, R. Bresin, Improving running mechanics by use of interactive sonification, in: Proceedings of the 3rd...
  • N. Schaffert et al.

    Designing an acoustic feedback system for on-water rowing training

    Int. J. Comput. Sci. Sport

    (2011)
  • Philips Research,...
  • Intersense,...
  • C. Brigante et al.

    Towards miniaturization of a MEMS-based wearable motion capture system

    IEEE Trans. Ind. Electron.

    (2011)
  • Cited by (20)

    • D-SORM: A digital solution for remote monitoring based on the attitude of wearable devices

      2021, Computer Methods and Programs in Biomedicine
      Citation Excerpt :

      Jamwal et al. [11] developed a wearable ankle robot to ensure remote rehabilitation programs, with a library of exercises creating virtual reality. Brückner et al. [3] presented an onboard inertial sensor fusion approach for home based stroke rehabilitation. A wearable platform was proposed, featuring an interface and a sound synthesis toolkit linked to wireless hearing aids.

    • Conditional equivalence between Extended Kalman filter and complementary filter for two-vector gyro-aided attitude determination

      2021, Measurement: Journal of the International Measurement Confederation
      Citation Excerpt :

      There are several weight regulation methods that have been proposed, which have summarized in [28,29]. There still do not exist a clear conclusion about which is better concerning the performances of EKF and CF. [30] gives two examples to illustrate that the time-invariant continuous linear Kalman filter is identical in form to the complementary filter, which can be seen as a preliminary work for performance comparison between Kalman filter and CF. [31,32] conclude by experiments that EKF behaves better than CF concerning attitude estimation accuracy, while it behaves poorer than CF concerning computational cost. The experimental results given by [13] conclude that there do not exist much performance difference between EKF and CF. The experimental results given by [33] conclude that CF has better accuracy than EKF.

    • Data fusion for orientation sensing in wireless body area sensor networks using smart phones

      2016, Pervasive Computing: Next Generation Platforms for Intelligent Data Collection
    View all citing articles on Scopus
    View full text