Physical human locomotion prediction using manifold regularization.

Area of Science:

• Biomechanics and Human Motion Analysis
• Signal Processing and Machine Learning

Background:

• Predicting human motion is crucial for research, but complex locomotion patterns remain challenging.
• Existing methods effectively recognize simple motion patterns but struggle with intricate dynamics.

Purpose of the Study:

• To propose novel methods for predicting complex human locomotion activities.
• To enhance the accuracy of human motion prediction using fused signals.

Main Methods:

• A calibration-based filter algorithm and kinematic-static patterns identification were developed.
• Inertial and physiological signals were pre-processed using specialized filters.
• Polynomial probability distribution and manifold regularization were employed for pattern analysis and performance assessment.

Main Results:

• The proposed system achieved 82.50% accuracy on the Physical Action Dataset (PAD).
• An accuracy rate of 81.90% was achieved on the Growing Old Together Validation (GOTOV) dataset.
• The developed prediction system demonstrated superior performance compared to state-of-the-art models.

Conclusions:

• The novel methods effectively predict complex human locomotion patterns from fused signals.
• The proposed system offers a significant advancement in human motion prediction accuracy.
• This research provides a robust framework for future studies in human dynamics.