Fusion of physically-based registration and deformation modeling for nonrigid motion analysis

Area of Science:

• Computer Vision
• Biomedical Engineering
• Mechanical Engineering

Background:

• Finite element models (FEM) are used for nonrigid motion analysis.
• Previous methods relied on separate tracking models and FEM, requiring correspondence data.
• Challenges include accurate registration, deformation modeling, and handling noise.

Purpose of the Study:

• To present a novel multiscale approach for nonrigid motion recovery from registered intensity and range images.
• To integrate registration and deformation modeling within a single FEM framework.
• To evaluate the method's speed, accuracy, and noise sensitivity.

Main Methods:

• A multiscale iterative algorithm utilizing undirected Hausdorff distance for correspondence recovery.
• A unified finite element model (FEM) incorporating material properties for registration and deformation.
• Demonstration on man-made elastic materials and human skin motion data.

Main Results:

• The proposed multiscale FEM approach achieves accurate nonrigid motion recovery.
• The method demonstrates robustness to noise and outperforms conventional separate models.
• It eliminates the need for a correspondence template, adapting to object features.

Conclusions:

• The novel multiscale FEM approach offers an effective strategy for nonrigid motion analysis.
• The method shows significant advantages in speed and accuracy over traditional techniques.
• Applications extend to motion tracking, analysis, and medical imaging, such as burn scar detection.