Identify dominant dimensions of 3D hand shapes using statistical shape model and deep neural network

Area of Science:

• Ergonomics and Human Factors
• Anthropometry
• 3D Shape Analysis

Background:

• Hand anthropometry is fundamental to ergonomic research and product design.
• Previous studies often lacked standardized hand dimension definitions, relying on researcher experience.
• Limited research explored the impact of individual hand dimensions on overall 3D hand shape variability.

Purpose of the Study:

• To identify dominant hand dimensions influencing hand shape variability.
• To consider the practical stability of these measurements.
• To establish a reliable set of dimensions for ergonomic applications.

Main Methods:

• Defined 58 landmarks and 53 dimensions from literature review.
• Utilized 80,000 virtual hand models and deep neural networks (DNNs) with statistical shape models (SSMs).
• Applied structured sparsity learning to identify dominant dimensions and evaluated measurement stability through manual methods.

Main Results:

• Identified 21 dominant dimensions capturing 90% of hand shape variance.
• Selected 16 dimensions with lower measurement variance based on stability and dominance.
• Developed a method to generate 3D hand models with 5.9 mm accuracy using these 16 dimensions.

Conclusions:

• The 21 identified dimensions serve as valuable references for anthropometric studies.
• The 16 selected dimensions enable efficient 3D hand model generation with minimal effort.
• This approach offers a practical, albeit limited, accuracy for early-stage ergonomic research and design.