A neural network approach to movement pattern analysis

Area of Science:

• Biomechanics
• Computational Neuroscience
• Sports Science

Background:

• Human movements are complex, time-dependent processes represented by high-dimensional coordinate data.
• Analyzing these movements traditionally requires extensive data and complex modeling.

Purpose of the Study:

• To explore the application of neural networks for analyzing movement patterns.
• To demonstrate dimensionality reduction from high-dimensional configurations to 2D trajectories.

Main Methods:

• Utilizing the Kohonen Feature Map (KFM), a type of neural network, to cluster movement patterns.
• Employing the Dynamically Controlled Network (DyCoN), a modified KFM, to reduce training data requirements.
• Modeling movements as time-series of coordinates, creating high-dimensional configurations.

Main Results:

• KFM successfully identifies patterns and clusters them, reducing dimensionality to 2D trajectories.
• DyCoN significantly reduces the need for extensive experimental training data.
• The method allows for automatic data transfer for further analysis and visualization.

Conclusions:

• Neural network-based analysis offers an effective method for movement pattern recognition and dimensionality reduction.
• DyCoN enhances the practicality of KFM for movement analysis by reducing data dependency.
• This approach supports, rather than replaces, expert analysis and model development in sports science.