ScholarWorks@숙명여자대학교

초록

A novel approach to wearable motion tracking that redefines sensor placement strategies is presented. While strain sensors offer compelling advantages over camera-based systems, most existing methods still rely on intuition-driven placement and complex machine-learning models that require extensive data and often generalize poorly. Recent sensor placement optimization studies have attempted to address these limitations using feature selection or search-based methods, yet they remain constrained to fixed sensor arrays or task-specific models that do not evaluate placement quality within an end-to-end motion tracking framework. This approach overcomes these limitations by leveraging computational strain mapping of joint motion and a genetic algorithm guided directly by model performance to identify optimal sensor configurations that traditional heuristics overlook. The method reveals counterintuitive yet highly effective placements, achieving a 32% reduction in tracking error compared to heuristic layouts. Moreover, this computational framework automatically determines optimal prestrain values, resolving a well-known limitation in strain sensor deployment. This data-driven framework not only delivers superior tracking performance but also dramatically accelerates the sensor configuration process, completing in hours what would traditionally require extensive manual testing, thereby enhancing wearable-sensor design by improving accuracy, efficiency, and practicality for applications in rehabilitation, sports science, and human-computer interaction.

키워드