Scholarlink

<p>Motion classification algorithms constitute a fundamental technical framework for a wide range of practical applications, including smart home interaction, home safety monitoring, intelligent driving, and sports status monitoring. Furthermore, they underpin cutting-edge research in fields such as exoskeleton robotics, human gait analysis, and biomechanics. The speed, accuracy, and capability to identify motion transition phases are critical performance metrics. These factors directly determine a system’s practical utility within real-time human–machine interaction interfaces and its ultimate effectiveness in rehabilitation training. To address the gap in recognizing transition phases of motion classification in previous research, this paper proposed a Self-adjusting Weighted Intelligent Transition-aware Classification Hybrid Algorithm (SWITCH). SWITCH integrates the Extreme Gradient Boosting model (XGBoost) and the Lightweight Gradient Boosting Machine model (LightGBM), incorporating a Transition Detection Module and an Adaptive Weight Adjustment Module. This configuration enables facilitates effective identification of five motions—standing, walking, running, upstairs, downstairs along with their transition phases, while minimizing computational resource consumption. During the research, a new human model was constructed based on the standard proposed by International Organization of Standard (ISO) for human-related data and the earlier human body model. The Center of Gravity (COG) positions of each human segment were quantified, providing theoretical support for Initial Measure Unit (IMU) mounting. Human motion data were collected, preprocessed, and compared with standard motion data to validate its consistence. Extensive experimental validation demonstrated an average recognition accuracy of 96% ± 0.67% and an average classification latency of 70.3&#xa0;ms, confirming the efficiency and accuracy of the algorithm.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research of human motion classification algorithm with adaptive weight adjustment module

  • 期刊论文

  • 发表日期：

    2026年5月23日

查看全文
  • Junping Wei,
  • Chenxi Qu,
  • Yang Zhou,
  • Fenglei Li,
  • Shengguan Qu

摘要

Motion classification algorithms constitute a fundamental technical framework for a wide range of practical applications, including smart home interaction, home safety monitoring, intelligent driving, and sports status monitoring. Furthermore, they underpin cutting-edge research in fields such as exoskeleton robotics, human gait analysis, and biomechanics. The speed, accuracy, and capability to identify motion transition phases are critical performance metrics. These factors directly determine a system’s practical utility within real-time human–machine interaction interfaces and its ultimate effectiveness in rehabilitation training. To address the gap in recognizing transition phases of motion classification in previous research, this paper proposed a Self-adjusting Weighted Intelligent Transition-aware Classification Hybrid Algorithm (SWITCH). SWITCH integrates the Extreme Gradient Boosting model (XGBoost) and the Lightweight Gradient Boosting Machine model (LightGBM), incorporating a Transition Detection Module and an Adaptive Weight Adjustment Module. This configuration enables facilitates effective identification of five motions—standing, walking, running, upstairs, downstairs along with their transition phases, while minimizing computational resource consumption. During the research, a new human model was constructed based on the standard proposed by International Organization of Standard (ISO) for human-related data and the earlier human body model. The Center of Gravity (COG) positions of each human segment were quantified, providing theoretical support for Initial Measure Unit (IMU) mounting. Human motion data were collected, preprocessed, and compared with standard motion data to validate its consistence. Extensive experimental validation demonstrated an average recognition accuracy of 96% ± 0.67% and an average classification latency of 70.3 ms, confirming the efficiency and accuracy of the algorithm.