Enhanced Human Crawling Phase Recognition Based on Kinematic Synergies and Machine Learning.

Area of Science:

• Biomechanics
• Rehabilitation Engineering
• Machine Learning

Background:

• Hands-and-knees crawling is crucial for motor rehabilitation in children.
• Precise phase detection is needed to optimize assistive devices for crawling.
• Current research on human crawling phase detection is limited.

Purpose of the Study:

• To evaluate the effectiveness of multijoint kinematic synergy (KS) features for crawling phase detection.
• To compare the accuracy of KS features against traditional time-domain (TD) features.
• To explore the application of improved phase detection in exoskeleton control systems for rehabilitation.

Main Methods:

• Nine healthy adults crawled while motion and pressure data were collected using accelerometers and pressure sensors.
• Kinematic synergy features were extracted using singular value decomposition-based principal component analysis (PCA).
• Machine learning models (CART, KNN, ECOC-SVM) were trained to recognize crawling phases using both KS and TD features.

Main Results:

• The kinematic synergy-based method achieved an average accuracy of 89.37%, with ECOC-SVM reaching 94.20%.
• Traditional time-domain features resulted in lower accuracies, with the highest overall being 75.36% for CART.
• Kinematic synergy features significantly outperformed TD features across all tested machine learning models.

Conclusions:

• Multijoint kinematic synergy features offer superior accuracy for crawling phase recognition compared to traditional time-domain features.
• Accurate phase recognition enables better interpretation of user intent for exoskeleton control.
• This approach has the potential to significantly improve rehabilitation outcomes for individuals with motor impairments, such as cerebral palsy.