Moving Target Detection In Dynamic Texture Background

Moving target detection is a key problem in the field of video analysis. It’s widelyapplication in video conference, remote teaching, traffic violation detection, militaryreconnaissance and intelligent human-machine interaction. However, for complex dynamicbackground, e.g., strong light, rain, swaying trees, waves and other moving objects, itbecomes more difficult to well detect the moving targets by stochastic motion. Therefore,moving target detection in dynamic background has become a concern of challengingresearch subject. Thus, we are mainly focus on the different between dynamic texturebackground and moving target. This paper is organized as follows.1. This paper proposes a center-surround discriminant saliency based detection methodusing the stationary wavelet transform (SWT) and the generalized Gamma distribution (GΓD).It makes use of the fact that the properties of the moving target and the dynamic backgroundis different in the stationary wavelet domain, and can be described by the GΓD model. Thecenter-surround discriminant saliency, which is to measure the difference between the currentlocation and its surround, is determined with the symmetrized Kullback-Leibler distance ofthe GΓD models of the center and surround in the wavelet transform. Then the moving targetis detected based on the threshold value method. The experimental results demonstrate theeffectiveness of the proposed method.2. Moving target detection basd on the auto-regressive and moving average model(ARMA). It has a good performance in dynamic texture feature extraction. This paperproposed the method based on the ARMR modle and center-surround discriminant saliency.Firstly, we use the Gaussian low-pass filter in the image frames to get the different resolutionsequences. And then, we use the least squares to estimate the ARMA model parameters.Finally, we get the saliency map of moving target by the Kullback-Leibler distance. Theexperimental results demonstrate the superior performance in moving target detection basedon the dynamic texture background.