In-Sensor Movement Variability Tracking.

Area of Science:

• Biomechanical Engineering
• Sensor Technology
• Machine Learning

Background:

• Movement variability tracking is crucial for analyzing and correcting human motion.
• Existing methods often require significant computational resources and data storage.
• On-chip implementation offers potential for real-time, low-power motion analysis.

Purpose of the Study:

• To develop an ultra-low-power, low-footprint, and data-efficient movement variability detection algorithm.
• To implement this algorithm on-chip within an inertial sensor for autonomous operation.
• To provide quantitative feedback for movement correction across various applications.

Main Methods:

• An on-chip algorithm for movement variability detection using inertial sensor data.
• Training phase: automatic segmentation and generation of approximate gravity vector templates.
• Inference phase: weighted similarity metrics and heatmap generation for deviation analysis.

Main Results:

• Algorithm operates with under 6 seconds of training data, 7 kB memory, and 0.2 mA current.
• Achieves a temporal resolution of 0.5 seconds.
• Generates heatmaps indicating deviations from movement templates for actionable correction insights.

Conclusions:

• The developed algorithm offers a highly efficient solution for on-sensor movement variability tracking.
• Its low power and memory requirements make it suitable for embedded systems.
• Broad applicability in sports analysis, exercise monitoring, rehabilitation, and gait tracking is demonstrated.