Detection method of moving object with small displacement

In video sequences, it generally exists that some parts of moving object overlap between different frames, such as bend, jumping-in-place, waving hands and so on. Thus, the overlapped parts are often taken as background wrongly. In order to improve the integrality of moving objects, an algorithm based on background subtraction and edge extraction is proposed. At first, the combination of adaptive median filter (AMF) and mean filtering was used to extract the cleaner background, and the general location of moving target was obtained by background subtraction. The improved Sobel operator was used to extract the complete edge from the original image simultaneously, and then the two results were fused to obtain a clear moving object. The proposed method was tested on the Weizmann human action dataset. The experimental results show that the approach is simple, and the detected object is complete with less running time.