Moving Object Detection in Complex Scene Using Spatiotemporal Structured-Sparse RPCA.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Moving object detection is crucial for computer vision.
• Robust Principal Component Analysis (RPCA) is commonly used but struggles with dynamic backgrounds and illumination variations.
• Existing methods lose spatiotemporal structure due to independence assumptions in the sparse component.

Purpose of the Study:

• To develop a spatiotemporal structured sparse RPCA algorithm for robust moving object detection.
• To address the limitations of traditional RPCA in complex and dynamic environments.
• To incorporate spatiotemporal structure into the sparse component for improved object separation.

Main Methods:

• Proposed a spatiotemporal structured sparse RPCA algorithm incorporating spatial and temporal regularization via graph Laplacians.
• Constructed multi-feature graphs over superpixels and enforced sparse components as eigenvectors.
• Solved the novel objective function using linearized alternating direction method of multipliers (ADMM) for batch optimization and proposed an online algorithm for real-time applications.

Main Results:

• The proposed algorithm effectively separates moving objects even in challenging conditions like dynamic backgrounds, camera jitter, and illumination changes.
• Both batch and online optimization methods demonstrated superior performance.
• Experiments on six public datasets validated the effectiveness against state-of-the-art methods.

Conclusions:

• The novel spatiotemporal regularization approach significantly improves moving object detection accuracy.
• The algorithm provides a robust solution for real-world computer vision applications with complex scenes.
• The developed batch and online methods offer versatile solutions for both offline analysis and real-time processing.