[Research on gait recognition and prediction based on optimized machine learning algorithm].

Area of Science:

• Biomechanics
• Artificial Intelligence
• Rehabilitation Engineering

Background:

• Human gait exhibits individual differences and variability, complicating recognition and prediction.
• Accurate gait analysis is crucial for rehabilitation, prosthetic development, and understanding movement disorders.

Purpose of the Study:

• To develop an optimized gait recognition and prediction method using motion posture signals.
• To address challenges posed by individual differences and stride variability in human lower limb movement.
• To improve the accuracy of predicting human body posture changes during various activities.

Main Methods:

• Utilized an optimized gated recurrent unit (GRU) network algorithm enhanced by immune particle swarm optimization (IPSO).
• Collected human body posture change data from multiple subjects performing walking, squatting, and leg flexion/extension.
• Performed comparative analysis against recurrent neural network (RNN), long short-term memory (LSTM), and standard GRU models.

Main Results:

• The IPSO-optimized GRU model demonstrated superior accuracy in predicting gait and posture changes.
• Achieved a root mean square error (RMSE) of 10^-3 for flat-land walking and squatting.
• Attained an RMSE of 10^-2 for sitting leg flexion and extension, with R² values above 0.966 for all actions.

Conclusions:

• The optimized algorithm effectively predicts human posture changes, showing high accuracy across different movements.
• This approach offers a valuable tool for gait movement evaluation and trend prediction in rehabilitation.
• Findings support applications in artificial limb design and lower limb rehabilitation equipment, enhancing patient independence.