Movement Estimation Using Soft Sensors Based on Bi-LSTM and Two-Layer LSTM for Human Motion Capture

Area of Science:

• Human-Computer Interaction
• Biomedical Engineering
• Robotics

Background:

• Human motion capture is crucial for various applications.
• Existing methods like HTC VIVE capture hand motion but often miss full arm articulation.
• There is a need for accurate, sensor-efficient arm orientation estimation.

Purpose of the Study:

• To propose a framework for estimating single-arm orientations using soft sensors.
• To improve the accuracy and reduce sensor dependency in arm motion capture.
• To combine Bi-LSTM and two-layer LSTM for enhanced estimation.

Main Methods:

• Utilized HTC VIVE for hand position tracking.
• Developed a framework combining Bi-long short-term memory (Bi-LSTM) and two-layer LSTM.
• Employed Myo gesture-control armbands for ground truth orientation data.
• Analyzed contextual features of consecutive sensory arm movements.

Main Results:

• The proposed framework achieved an average of 73.90% less dynamic time warping distance compared to conventional Bayesian methods.
• Demonstrated accurate estimation of upper arm and forearm orientations.
• Successfully reduced the number of sensors required for end-users.

Conclusions:

• The framework offers an efficient and accurate method for single-arm orientation estimation.
• It enables arm orientation estimation with any soft sensor, ensuring good accuracy.
• The combination of Bi-LSTM and two-layer LSTM is a key contribution for improved motion capture.