A Robust Kalman Algorithm to Facilitate Human-Computer Interaction for People with Cerebral Palsy, Using a New Interface Based on Inertial Sensors

Rafael Raya, Eduardo Rocon, Juan Alvaro Gallego, Ramón Ceres Ruíz, José Luis Pons2012 · DOI

TL;DR

This paper introduces ENLAZA, a novel human-computer interface (HCI) for individuals with cerebral palsy (CP) that utilizes inertial sensors to characterize and filter involuntary movements. A robust Kalman filter (RKF) is proposed and validated, significantly improving cursor control and target acquisition speed for users who previously struggled with conventional interfaces.

What Problem Does This Paper Address?

Conventional human-computer interfaces (HCI) often present significant challenges for individuals with cerebral palsy (CP) due to their motor impairments. While various assistive technologies exist, there's a notable gap in solutions specifically addressing the severe motor limitations faced by the CP community. Involuntary movements, characteristic of CP, can severely disrupt fine motor control, making tasks like precise cursor manipulation difficult or impossible. This paper aims to bridge this gap by developing an HCI that can effectively characterize and mitigate the impact of these involuntary movements, thereby enhancing computer accessibility for individuals with CP.

Methods

The core of this research lies in the ENLAZA interface and a Robust Kalman Filter (RKF).

  • ENLAZA Interface: This system employs an Inertial Measurement Unit (IMU) comprising a three-axis gyroscope, accelerometer, and magnetometer. This IMU, integrated into a headset, captures head movements. The system translates sagittal and transverse head rotations into pointer movements on the screen.
  • Characterization of Motor Disorders: Before filtering, the paper details a method to characterize the motor disorders of CP users. This involves:
  • Time-domain analysis: Assessing the correlation between voluntary psychomotor models and captured data (R-squared) to understand speed-accuracy trade-offs.
  • Frequency-domain analysis: Using Fast Fourier Transform (FFT) to identify dominant frequencies in movements, with a low-pass filter at 10 Hz to remove electrical noise without affecting voluntary or involuntary movements.
  • Spatial-domain analysis: Measuring head orientation and Range of Motion (ROM) to assess postural control.
  • Robust Kalman Filter (RKF): This is the key filtering technique. It builds upon the standard Kalman filter by incorporating a Huber's function for robust estimation. This approach is designed to detect and downweight outlying observations (interpreted as involuntary sub-movements) within the least squares method, thereby improving the accuracy of the estimated cursor position. The RKF dynamically adapts its gain and is designed to be a bidirectional filter, handling independent vertical and horizontal effects.

Datasets

The paper does not explicitly mention the use of pre-existing datasets for training. Instead, it focuses on:

  • Data Collection from Participants: Data was collected from four individuals with severe CP (CP1, CP2, CP3, CP4) and three healthy users.
  • Experimental Protocols: Participants performed target-reaching tasks under various conditions (without filter, with BBF, KF, and RKF).
  • Benchmark Comparison: The performance of the ENLAZA interface and the RKF was compared against existing methods like the Benedict-Bordner Filter (BBF) and the standard Kalman Filter (KF), as well as an "incremental method" for filtering.

Key Insights and Ideas

  • Voluntary and Involuntary Movements Share Bandwidth: A crucial finding is that for individuals with CP, voluntary and involuntary movements often share the same frequency bandwidth. This implies that simple frequency-based filtering might not be sufficient and necessitates more sophisticated approaches that can differentiate between intended and unintended movements.
  • RKF's Superiority in Outlier Rejection: The RKF's ability to robustly handle outliers (involuntary sub-movements) proved to be more effective than traditional filters like BBF and KF for the specific challenges posed by CP motor disorders. This is attributed to its iterative downweighting of outlying observations.
  • ENLAZA's Potential for Severe Motor Impairments: The study demonstrates that the ENLAZA interface, coupled with the RKF, can enable computer control for individuals with CP who were previously unable to use conventional or even some advanced assistive interfaces.
  • Fine Motor Control Enhancement: The RKF significantly improves fine motor control by reducing the number of "sub-movements" around the target, leading to smoother cursor trajectories and more precise targeting.

Key Results

  • Reduced Reaching Time: The RKF algorithm significantly reduced target-reaching time for CP users. For CP1, the time was reduced by a factor of 10 (from 109s to 8.67s). For CP3 and CP4, the reduction was by a factor of 2 (from ~44s to ~18s and ~43s to ~17s, respectively).
  • Reduced Sub-movements: Filtering techniques, particularly the RKF, substantially reduced the number of involuntary sub-movements. The RKF reduced movement segmentation by up to 65%.
  • Improved Cursor Directivity and Target Acquisition: The paper states that the filter increases mouse pointer directivity and reduces target acquisition time by a factor of ten.
  • Validation with Non-Conventional Users: The ENLAZA interface and RKF enabled users with severe CP, who could not use other interfaces, to control the computer effectively.

Limitations and Future Work

  • Limited Number of Participants: The study involved a small number of participants with CP (four).
  • Short-Term Experiments: The experiments were conducted over a week, and long-term studies are needed to analyze the effects of physical and cognitive learning on device control.
  • Filtering Strategy Independence: The current filtering strategy is independent of target location. Future work could explore integrating this with techniques that adapt to the environment, such as "click crossing."
  • Application to Other Interfaces: The authors plan to apply the RKF to other alternative HCI, such as eye-tracking systems, to improve their accessibility for users with involuntary movements.
  • Broader User Groups: Extending the study to include more users with CP and other groups with similar motor disabilities (e.g., spinal cord injury, stroke) is suggested.
  • Full-Body Motion Tracking: The development of a full-body motion tracking system for analysis and therapy is envisaged.
  • Therapy Evaluation: Evaluating the impact of therapies using objective parameters and virtual representation via biofeedback is a future direction.

Summary updated

BibTeX
@misc{raya2012robust,
  author = {Rafael Raya and Eduardo Rocon and Juan Alvaro Gallego and Ramón Ceres Ruíz and José Luis Pons},
  title = {{A Robust Kalman Algorithm to Facilitate Human-Computer Interaction for People with Cerebral Palsy, Using a New Interface Based on Inertial Sensors}},
  year = {2012},
  doi = {10.3390/s120303049},
}
Figure