A Structure Constraint Matrix Factorization Framework for Human Behavior Segmentation.

Area of Science:

• Computer Vision
• Machine Learning
• Data Science

Background:

• Human behavior segmentation from sequential data is challenging due to high dimensionality and similarity.
• Existing methods struggle with precise segmentation of continuous actions.

Purpose of the Study:

• To present a structure constraint matrix factorization framework for accurate human behavior segmentation.
• To address the limitations of current algorithms in handling complex sequential behavior data.

Main Methods:

• Developed a framework based on behavior continuity and high similarity between frames.
• Constructed effective constraint regular terms using the behavior continuity hypothesis.
• Established a clustering framework using constrained non-negative matrix factorization, spectral clustering, and graph segmentation.

Main Results:

• The framework successfully segmented human behavior sequences across multiple public datasets (Weiz, Keck, mo_86, mo_86_9).
• Demonstrated automatic and accurate segmentation of human behavior.
• Achieved consistent segmentation of sequential points within behavior actions.

Conclusions:

• The proposed structure constraint matrix factorization framework offers superior performance in human behavior segmentation accuracy.
• This approach provides a robust solution for analyzing sequential human behavior data.