Moving object extraction using a basis function network

Abstract

This paper describes a moving object extraction method using a basis function network. The basis function network represents frame images as a linear conjunction of weighted basis functions. The ordinary background subtraction method can't extract the moving object from unstable background. The ordinary frame differencing method can't extract the whole moving object. The proposed method can detect a moving object by the difference between consecutive weights and can reconstruct a local image including a whole moving object. Detections and extractions of moving objects from unstable background are examined and the validity of the proposed method is discussed.