Self-Motion-Assisted Tensor Completion Method for Background Initialization in Complex Video Sequences

Area of Science:

• Computer Vision
• Image and Video Processing
• Machine Learning

Background:

• Background Initialization (BI) is crucial for video analysis.
• Spatiotemporal slice-based singular value decomposition (SS-SVD) is a tensor-based BI method.
• SS-SVD struggles with complex sequences (e.g., stationary foreground objects, illumination changes, low frame-rate).

Purpose of the Study:

• To propose a novel self-motion-assisted tensor completion method for enhanced background initialization.
• To overcome the limitations of SS-SVD in complex video scenarios.
• To improve the visual quality of initialized backgrounds.

Main Methods:

• A self-motion-assisted tensor completion technique is introduced.
• Motion information from sparse tensor slices is integrated with SS-SVD's low-rank information.
• Efficient blending schemes are developed for low-rank (background) and sparse (foreground) data.

Main Results:

• The proposed method effectively eliminates artifacts in initialized backgrounds.
• It demonstrates improved performance in handling stationary foreground objects, illumination variations, low frame-rate, and clutter.
• The method achieves superior results compared to state-of-the-art techniques in complex scenarios.
• It also reduces computational time for background initialization.

Conclusions:

• The self-motion-assisted tensor completion method significantly enhances background initialization in challenging video sequences.
• It offers a robust and efficient solution for various video processing applications.
• The technique provides better visual quality and handles complex dynamic scenes more effectively.