Generic Content-Based Retrieval of Marker-Based Motion Capture Data.

Area of Science:

• Computer Vision
• Biomechanical Engineering
• Data Science

Background:

• Marker-based motion capture is widely used in animation, biomechanics, and robotics.
• Existing retrieval methods often struggle with data variability in subject types and marker conventions.
• A need exists for a robust and generic content-based retrieval system for motion capture data.

Purpose of the Study:

• To propose an original scheme for generic content-based retrieval of marker-based motion capture data.
• To develop a novel motion signature capable of describing both high-level and low-level motion characteristics.
• To enable retrieval across arbitrary subject types and marker labeling conventions.

Main Methods:

• A novel motion signature is proposed to statistically describe morphological and kinematic features of motion sequences.
• Content-based retrieval is performed by calculating the motion signature distance between a query and database items.
• Optimal weighting schemes for motion signatures are pre-learned using biased discriminant analysis and adapted at runtime.

Main Results:

• The proposed scheme demonstrates excellent performance in content-based retrieval of motion capture data.
• Experimental validation on various datasets and metrics confirms the effectiveness of the novel motion signature.
• The method proves effective for arbitrary subject types and marker attachment conventions.

Conclusions:

• The developed scheme offers a generic and effective solution for content-based retrieval of marker-based motion capture data.
• The novel motion signature and adaptive weighting system significantly enhance retrieval performance.
• This work advances the field by providing a flexible tool for managing and searching large motion capture databases.