Improved Gaussian Mixture Model with Background Spotter for the

Extraction of Moving Objects

Brahim Farou1,2, Hamid Seridi2, and Herman Akdag3

1Computer Science Department, Badji Mokhtar-Annaba University, Algeria

2LabSTIC, Guelma University, Algeria

3LIASD, Paris 8 University, France

 Abstract: Extraction of moving objects is a key step in a visual surveillance area. Many background models have been proposed to resolve this problem, but Gaussian Mixture Model (GMM) remains the most successful approach for background subtraction. However, the method suffers from sensitivity (SE) to local variations; variations in the brightness and background complexity mislead the process to a false detection. In this paper, an efficient method is presented to deal with GMM problems through improvement on updating selected pixels by introducing a background spotter. First, the extracted frame is divided into several equal size regions. Each region is assigned to a spotter who will report significant environment changes based on histogram analysis. Only parts reported by spotters are considered and updated in the background model. Tests carried out on four video databases that take into account various factors, demonstrate the effectiveness of our system in real-world situations.

 Keywords: Video surveillance, GMM, modeling the background, image processing.