An LSTM-Based Prediction Method for Lower Limb Intention Perception by Integrative Analysis of Kinect Visual Signal.

Area of Science:

• Biomedical Engineering
• Rehabilitation Robotics
• Computer Vision in Healthcare

Background:

• Gait rehabilitation research increasingly utilizes computer vision and deep learning.
• Long Short-Term Memory (LSTM) networks excel at learning sequential data representations.
• Accurate prediction of limb movement is crucial for effective robotic-assisted rehabilitation.

Purpose of the Study:

• To propose a novel lower limb joint trajectory prediction method for active rehabilitation.
• To leverage LSTM networks and upper limb motion data for predicting lower limb intentions.
• To validate the feasibility of using Kinect visual signals for gait rehabilitation.

Main Methods:

• Developed a custom Kinect-Treadmill data acquisition platform.
• Collected upper and lower limb joint angle data from ten healthy subjects at four walking speeds.
• Trained an LSTM model to predict contralateral lower limb joint angles (hip, knee) from ipsilateral upper limb joint angles (elbow, shoulder).

Main Results:

• The trained LSTM model demonstrated accurate estimation of lower limb intentions.
• Predicted joint trajectories showed a strong correlation with actual limb movements.
• The study validated the effectiveness of Kinect visual signals in capturing relevant motion data for rehabilitation.

Conclusions:

• The proposed LSTM-based method effectively predicts lower limb joint trajectories for robotic gait rehabilitation.
• Synergy theory supports the use of upper limb motion to infer lower limb intentions.
• Kinect technology is a feasible tool for acquiring motion data in gait rehabilitation research.