Moving Object Detection by Detecting Contiguous Outliers in the Low-Rank Representation

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Automated video analysis relies heavily on object detection.
• Traditional methods like object detectors and background subtraction require extensive manual labeling or specific training sequences.
• Existing motion-based approaches struggle with complex scenarios like non-rigid motion and dynamic backgrounds.

Purpose of the Study:

• To develop a unified framework for automated object detection without a separate training phase.
• To address the limitations of current methods in handling complex video analysis scenarios.
• To introduce a novel approach that integrates object detection and background learning.

Main Methods:

• A unified framework named DEtecting Contiguous Outliers in the LOw-rank Representation (DECOLOR) was developed.
• The DECOLOR formulation integrates object detection and background learning into a single optimization process.
• An alternating algorithm was employed for efficient solution of the optimization problem.

Main Results:

• DECOLOR effectively addresses challenges posed by non-rigid motion and dynamic backgrounds.
• The framework integrates object detection and background learning seamlessly.
• Experiments show DECOLOR outperforms state-of-the-art approaches on simulated and real-world data.

Conclusions:

• DECOLOR provides an efficient and effective solution for automated object detection in complex video scenarios.
• The unified framework demonstrates superior performance compared to existing methods.
• DECOLOR offers a robust approach for diverse computer vision applications.