Spatiotemporal GMM for Background Subtraction with Superpixel Hierarchy.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Background subtraction is crucial for video analysis.
• Existing methods struggle with dynamic scenes and sudden pixel value changes.

Purpose of the Study:

• To develop a robust background subtraction algorithm.
• To improve performance in videos with frequent and sudden pixel value changes.

Main Methods:

• Hierarchical superpixel segmentation.
• Construction of spanning trees from Gaussian Mixture Models (GMMs).
• Application of the -smoother and optical flow for spatio-temporal consistency.

Main Results:

• The proposed algorithm demonstrates favorable performance on synthetic and real-world datasets.
• It effectively handles videos with frequent and sudden pixel value changes.
• Outperforms state-of-the-art background subtraction methods.

Conclusions:

• The novel algorithm offers a robust solution for background subtraction.
• It shows significant improvements in challenging video scenarios.
• Provides a valuable contribution to video analysis research.