Dual-Branch Deep Learning for Continuous Gait Cycle Estimation with wearable IMU Sensors and Anthropometric Data

Abstract

This study presents a CNN-LSTM model for gait cycle phase prediction using wearable IMU data and anthropometric features. A sliding window approach was applied for real-time gait estimation, with gait cycle percentage calculated via peak detection and lower limb length data. Feature selection identified key IMU-derived kinematic features, primarily joint angles, to enhance model performance. The model achieved R² = 0.911, demonstrating high accuracy across walking conditions. These findings support IMU-based gait analysis for rehabilitation, abnormal gait detection, and prosthetic control. To achieve our ultimate goal in the future, we are developing a comprehensive AI-driven system for real-time gait cycle assessment.Clinical relevance - This study supports real-time gait monitoring using wearable IMU sensors, enabling personalized rehabilitation, early abnormal gait detection, and prosthetic adaptation. The model's accuracy and reliance on kinematic data alone make it suitable for clinical and home-based gait assessment.