Domain-Transformable Sparse Representation for Anomaly Detection in Moving-Camera Videos.

Area of Science:

• Computer Vision
• Machine Learning
• Signal Processing

Background:

• Anomaly detection in video sequences is challenging, especially with moving cameras.
• Existing methods struggle with camera motion and visual clutter.

Purpose of the Study:

• To develop a robust anomaly detection system for video sequences with moving cameras.
• To improve geometric registration between reference and target videos for better anomaly identification.

Main Methods:

• A sparse representation-based matrix factorization technique is employed.
• Domain transformations are incorporated to handle camera motion.
• Approximations are used to create a feasible iterative optimization process.

Main Results:

• The proposed algorithm achieves superior geometric registration between videos.
• Significant improvements in overall anomaly detection performance are demonstrated.
• Effective handling of camera trepidations and visually cluttered environments.

Conclusions:

• The novel matrix factorization method enhances anomaly detection in challenging video scenarios.
• The technique offers a robust solution for real-world applications involving moving cameras.