Real-time strength training guidance system
using automatically generated evaluation rules

Abstract

With the growing health consciousness, home-based training has become widespread. However, without professional guidance, risks such as injuries from improper form and reduced exercise effectiveness remain significant challenges. This study proposes a real-time feedback system for strength training that learns from user-provided reference videos. By using "optimal rep" data extracted from videos as a baseline, we realized a real-time feedback system that combines LLM-based evaluation rule generation with synchronization using Dynamic Time Warping (DTW). This approach provides a personalized training experience that addresses both "exercise diversity" and "real-time performance," which were difficult to achieve with conventional rule-based methods.