An effective identification method of video ships and vehicles

Abstract

In usual videos, there are problems such as small amplitude motion interference and sudden change of illumination in the background. It is difficult for ship and vehicle recognition. This paper takes full advantage of classical mixture Gauss background model in the context of small-scale motion. The inter frame difference method is not sensitive to illumination changes. So this paper proposes an identification method combining mixture of Gauss background model and three-frame difference method, and achieves potential region acquisition. The connected components are analyzed and merged to obtain the minimum outer rectangles, and the geometric characteristics of the targets are extracted from the rectangles. This paper uses the LS_SVM classifier to classify each potential region and outputs whether there are targets and the location of the targets. The experiment results show the average accuracy of the target recognition method in this paper is significantly improved.