Learning and removing cast shadows through a multidistribution approach

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Moving cast shadows pose a significant challenge for foreground detection in surveillance.
• Accurate shadow identification and removal are crucial for reliable foreground image processing.

Purpose of the Study:

• To develop a novel pixel-based statistical approach for modeling moving cast shadows with nonuniform and varying intensity.
• To enhance foreground detection accuracy in surveillance applications by effectively handling shadows.

Main Methods:

• Utilized Gaussian mixture model (GMM) learning to create statistical models of moving cast shadows on surfaces.
• Developed a pixel-based approach capable of handling complex, time-varying illumination and light-saturated areas.
• Integrated the proposed method with existing pixel-based shadow descriptions.

Main Results:

• The approach effectively models shadows with nonuniform and varying intensity.
• Significantly reduced the false detection rate of shadows without increasing the missed detection rate.
• Demonstrated robustness across various scene types and shadow models.

Conclusions:

• The proposed GMM-based statistical modeling offers a robust solution for moving cast shadow detection.
• This method improves the reliability of foreground detection in surveillance systems under challenging lighting conditions.
• The approach successfully prevents false detections in areas where shadows are not present.