A Foreground Extraction Algorithm Based on Adaptively Adjusted Gaussian Mixture Models

Background subtraction is a widely used method for moving object detection in computer vision field. To cope with highly dynamic and complex environments, the mixture of models has been proposed. In this paper, a background subtraction method is proposed based on the popular Gaussian Mixture Models technique and a scheme is put forward to adaptively adjust the number of Gaussian distributions aiming at speeding up execution. Moreover, edge-based image is utilized to weaken the effect of illumination changes and shadows of moving objects. The final foreground mask is extracted by the proposed data fusion scheme. Experimental results validate the performance of proposed algorithm in both computational complexity and segmentation quality.