Student's t-distribution mixture background model for efficient object detection
Ling Hua Guo, Minghui Du · 2012
Background subtraction is an essential technique for moving object segmentation in vision surveillance system. To acquire an exact background, Gaussian mixture modeling (GMM) is a popular method for its adaptation to background variations. However, limited training samples and complex scenes result in heavy tails for GMM, which significantly affect the moving object detection accuracy. By reviewing the formulations of GMM, we construct a student's t-distribution mixture background model (SMBM) on the basis of fuzzy c-means clustering partition algorithm. Then, we present a method for moving object segmentation based on confidence analysis. Experimental results show that the background model can reflect complex scenes; our method achieves efficient object detection than conventional GMM approaches.