Fusing Self-Organized Neural Network and Keypoint Clustering for Localized Real-Time Background Subtraction

Area of Science:

• Computer Vision
• Artificial Intelligence
• Video Analytics

Background:

• Moving object detection is crucial for tasks like object classification and tracking.
• Dynamic aspects such as bootstrapping and illumination changes pose challenges.
• Pan-Tilt-Zoom (PTZ) cameras introduce further complexity due to mixed movements.

Purpose of the Study:

• To propose a real-time moving object detection method for PTZ camera video sequences.
• To address challenges in dynamic aspects and PTZ camera movements.
• To enhance foreground detection accuracy and robustness.

Main Methods:

• A combined keypoint clustering and neural background subtraction approach.
• Utilizing a Self-Organized Neural Network (SONN) for background subtraction.
• Spatio-temporal tracking of keypoints for foreground/background recognition.

Main Results:

• The method effectively detects moving objects in video streams.
• It demonstrates proficiency in managing bootstrapping and illumination variations.
• Experimental results validate efficiency in background modeling and subtraction.

Conclusions:

• The proposed method offers an efficient solution for moving object detection with PTZ cameras.
• It successfully handles dynamic environmental changes.
• The approach contributes to advancements in real-time video analysis.