LOW IMU SAMPLING RATES BIAS LARGEST LYAPUNOV EXPONENT CALCULATIONS DURING OVERGROUND WALKING
Presenter Type
UNO Graduate Student (Doctoral)
Major/Field of Study
Biomechanics
Author ORCID Identifier
0000-0001-9915-6950
Advisor Information
Assistant Professor
Location
MBSC Ballroom Poster # 1002 - G (Doctoral)
Presentation Type
Exhibit
Start Date
24-3-2023 9:00 AM
End Date
24-3-2023 10:15 AM
Abstract
Human activity is variable – movements vary slightly from one cycle to the next. These deviations can be quantified and provide information about the health of the neuromuscular system over time. Creating guidelines for studying movement variability requires deep knowledge about the performance of analytical tools in various data conditions such high vs. low sampling rates (SRs). The effects of changing the number of samples per second using optical motion capture are well known, but less is known about those effects on data collected from inertial measurement units (IMUs). IMUs are small and portable sensors that can record many movements such as walking. We expected that decreasing SRs would change the estimation of the Largest Lyapunov Exponent, λ1, a common variability metric. The λ1 measures how different the trajectory of a moving system is after each cycle. For example, if you were to draw a circle many times, λ1 will inform you how much each line changes from the last as you complete each moving cycle. 35 young adults (630 trials, total) walked for four minutes overground on a 200 m oval indoor track. Participants wore 16 IMUs (SR=200 Hz) that captured whole body movement. Manufacturer recommended corrections (course stabilization, fusion models, and anti-wobbling corrections) were applied before analysis. Segment angles of the thigh, shank, and foot were down sampled to 175, 150, 125, 100, 75, 50, and 25 Hz before computing λ1. Two common algorithms (Wolf and Rosenstein) were calculated for λ1 for the thigh, shank, and foot segment angles. A series of 1 (sensor) x 8 (SR) RM-ANOVAs were performed to test whether λ1 varied as a function of SR. Reducing the SR of our data resulted in significant λ1 differences when calculated by Wolf and Rosenstein at all sensor locations. Furthermore, as SR decreases, the λ1 trends positively when calculated with Rosenstein but remains stable with Wolf. 25 Hz is a breaking point where neither algorithm accurately estimates λ1 in this dataset. This study is funded by NSF 212491, NIH P20GM109090, R01NS114282, University of Nebraska Collaboration Initiative, the Center for Research in Human Movement Variability.
Scheduling
9:15-10:30 a.m., 10:45 a.m.-Noon, 1-2:15 p.m.
LOW IMU SAMPLING RATES BIAS LARGEST LYAPUNOV EXPONENT CALCULATIONS DURING OVERGROUND WALKING
MBSC Ballroom Poster # 1002 - G (Doctoral)
Human activity is variable – movements vary slightly from one cycle to the next. These deviations can be quantified and provide information about the health of the neuromuscular system over time. Creating guidelines for studying movement variability requires deep knowledge about the performance of analytical tools in various data conditions such high vs. low sampling rates (SRs). The effects of changing the number of samples per second using optical motion capture are well known, but less is known about those effects on data collected from inertial measurement units (IMUs). IMUs are small and portable sensors that can record many movements such as walking. We expected that decreasing SRs would change the estimation of the Largest Lyapunov Exponent, λ1, a common variability metric. The λ1 measures how different the trajectory of a moving system is after each cycle. For example, if you were to draw a circle many times, λ1 will inform you how much each line changes from the last as you complete each moving cycle. 35 young adults (630 trials, total) walked for four minutes overground on a 200 m oval indoor track. Participants wore 16 IMUs (SR=200 Hz) that captured whole body movement. Manufacturer recommended corrections (course stabilization, fusion models, and anti-wobbling corrections) were applied before analysis. Segment angles of the thigh, shank, and foot were down sampled to 175, 150, 125, 100, 75, 50, and 25 Hz before computing λ1. Two common algorithms (Wolf and Rosenstein) were calculated for λ1 for the thigh, shank, and foot segment angles. A series of 1 (sensor) x 8 (SR) RM-ANOVAs were performed to test whether λ1 varied as a function of SR. Reducing the SR of our data resulted in significant λ1 differences when calculated by Wolf and Rosenstein at all sensor locations. Furthermore, as SR decreases, the λ1 trends positively when calculated with Rosenstein but remains stable with Wolf. 25 Hz is a breaking point where neither algorithm accurately estimates λ1 in this dataset. This study is funded by NSF 212491, NIH P20GM109090, R01NS114282, University of Nebraska Collaboration Initiative, the Center for Research in Human Movement Variability.