Estimating Muscle Activity from the Deformation of a Sequential 3D Point Cloud

Area of Science:

• Biomechanics
• Computer Vision
• Machine Learning

Background:

• Estimating muscle activity is crucial for understanding human movement and intentions.
• Non-contact methods, like visual measurement systems, offer potential for accessible muscle activity data.
• Accurate 3D reconstruction and motion tracking are vital for reliable surface-based biomechanical analysis.

Purpose of the Study:

• To develop a method for predicting human muscle activity from skin surface strain.
• To address challenges in 3D reconstruction accuracy caused by noise in visual measurement systems.
• To improve the precision of human skin surface shape and motion tracking over time.

Main Methods:

• Proposed a method incorporating temporal constraints into non-rigid registration for accurate point cloud motion tracking.
• Developed a multilayer perceptron artificial neural network to infer muscle activity using surface strain as input.
• Trained and validated the network using data from simple lower limb movements.

Main Results:

• Successfully achieved accurate tracking of human skin surface shape and motion.
• Demonstrated the capability to estimate muscle activity directly from surface strain data.
• The proposed method effectively overcomes noise-induced reconstruction errors in visual measurement systems.

Conclusions:

• The developed method enables reliable, non-contact estimation of muscle activity via skin surface strain.
• Temporal constraints in non-rigid registration significantly improve motion tracking accuracy.
• This approach provides a valuable tool for biomechanical studies and movement analysis.