Analysis Of The Application In Moving Target Detection Based On Feature Extraction

Moving target detection is an important fundamental work in computer visionapplications, and it will effectively affect the subsequent applications such astracking, target identification, and action recognition. It also has a wide range ofapplications on the intelligent monitoring system, intelligent traffic system,human-computer interaction and content based video coding. However, theenvironment of reality is complex and always changing. The illumination variations,dynamic background, projection shadow and random noise have increased thedifficulty of target detection. At present, researchers have proposed many detectionalgorithms, this paper summarize and analyzed the most representative methodincluding frame difference method, optical flow method and background modelingmethod.Foreground segmentation is the last step in target detection, the traditional pixellevel classification and block-based classification method both have theirshortcomings. Pixel level foreground segmentation ignore the rich neighborhoodspatial information, and block-based classification may exhibit "blockiness" artifacts.In view of this, this paper introduces a new method based on the overlapping block,this method can overcome the shortage of pixel classification and block-basedsegmentation method, and also has a self correcting mechanism which caneffectively reduce the probability of foreground pixels misclassified.On the basis of previous research, the main contributions of this paper are asfollows:(1)this paper introduced the principle of moving target detection method, andthen analyzed several classical moving target detection algorithms, elaborate thedefinition and updating of Gauss mixture model.(2)We learned the modeling and updating strategy of2D PCA, proposed amoving target detection method which combined2D PCA and the overlapping blockmethod.(3)We learned the universe local binary pattern, proposed a moving targetdetection algorithms which combined LBP histogram and overlapping block method.The proposed method can reduce the number of histogram and restrain noise effectively.(4)The proposed algorithms were implemented in C++with the aid of OpenCVlibraries on Visual Studio2010platform. The experiment results were comparedwith the ground-truth foreground masks to analyse the proposed algorithms.