The detection of moving objects in video by background subtraction using Dempster-Shafer theory [articol]

Abstract

Detection of moving objects has been widely used in many computer vision applications like video surveillance, multimedia applications, optical motion capture and video object segmentation. The key steps in detecting the moving objects are the background subtraction and the foreground detection. To handle these processes, we need to classify the corresponding pixels of the current image as background or foreground. This paper describes the background subtraction and the foreground detection within the context of Dempster-Shafer theory which better represents uncertainty by considering the situations of risk and ignorance. The proposed method addresses the methodology modeling in the Dempster-Shafer theory of evidence by representing the information extracted from the current image as measures of belief. The mass functions are computed from the probabilities assigned to each class being combined with the Dempster-Shafer rule of combination and the maximum of mass function is used for decision-making. The proposed method has been tested on several datasets showing an optimal performance compared to other fuzzy approaches based on the Sugeno and Choquet integrals and has proved its robustness.