Trade-Offs Between Simplifying Inertial Measurement Unit-Based Movement Recordings and the Attainability of Different...

Area of Science:

• Biomechanics and Human Movement Analysis
• Wearable Technology and Sensor Networks
• Human-Computer Interaction and Behavioral Science

Background:

• Human movement activity is frequently monitored using inertial measurement unit (IMU) sensors across various scientific fields.
• IMU data enables algorithmic detection of postures and movements, aiding in detailed assessments of complex daily behaviors.
• Understanding the trade-offs between IMU recording configurations and analytical outcomes is essential for real-world behavioral studies.

Purpose of the Study:

• To systematically evaluate the impact of IMU recording configurations on posture and movement detection.
• To assess the effects of sensor number, placement, sampling frequency, and modality on high and low temporal resolution movement analyses.
• To determine optimal IMU configurations for naturalistic daily activity analysis without repetitive movements.

Main Methods:

• Utilized a dataset from 41 infants (4-18 months) wearing a multisensor suit for naturalistic movement recording.
• Established a benchmark using human annotations from synchronized video recordings.
• Systematically varied IMU sensor placement, number, sampling frequency (down to 6 Hz), and modality (accelerometer/gyroscope) to assess classification performance for 7 postures and 9 movements.

Main Results:

• Reducing the number of IMU sensors severely impacted classifier performance, rendering single-sensor configurations infeasible.
• Using accelerometer-only data resulted in a modest decrease in movement classification accuracy.
• Sampling frequencies could be reduced to 6 Hz with negligible impact on posture and movement classification accuracy.

Conclusions:

• Significant trade-offs exist between IMU recording configurations and the reliability of movement analyses.
• Single-sensor IMU setups are of limited utility for assessing real-world movement behavior.
• A minimal effective configuration includes upper/lower extremity sensors, at least 13 Hz sampling, and accelerometer/gyroscope data for reliable posture and movement quantification.