Moving object detection for video surveillance.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Signal Processing

Background:

• Independent living is increasingly supported by video surveillance systems.
• Existing video surveillance algorithms struggle with environmental challenges like illumination variations, appearance changes, background clutter, and occlusions.
• Robust object detection remains a critical challenge in developing effective video surveillance.

Purpose of the Study:

• To present a novel object detection algorithm for enhanced video surveillance.
• To improve the robustness of object detection in home-based independent living scenarios.
• To address limitations of current surveillance algorithms by incorporating advanced image processing techniques.

Main Methods:

• Video compression using two-dimensional discrete cosine transform (2D DCT) to reduce storage.
• Object detection via statistical correlation for key feature point identification.
• Classification of feature points into foreground and background using Bayesian rule.
• Object localization by embedding maximum likelihood feature points in successive frames.

Main Results:

• The proposed algorithm effectively detects and localizes objects in video surveillance.
• Experimental results demonstrate the algorithm's effectiveness against various surveillance metrics.
• Comparative studies confirm the superiority of the novel approach over existing methods.

Conclusions:

• The developed object detection algorithm offers a robust solution for video surveillance.
• The approach effectively handles common challenges in surveillance, improving reliability.
• This method enhances the potential of video surveillance for supporting independent living.