Studies On Camouflaged Moving Objects Detection In Video With Complex Background

Moving object detection is an essential technology in computer vision, and plays an important role in intelligent traffic, robot vision navigation, intelligent surveillance and military field, et al. Moving object detection, feature extraction and identification, are the key technologies in video surveillance. Moving target detection is the basis for target tracking, the test results are related to the accuracy of target tracking directly. Therefore, there is great theoretical and application value in researching the detection moving objects.This paper mainly studies pixel labeled algorithm of connecting area in binary images, the gray value change of camouflaged moving objects in video, detection of camouflaged moving objects in dynamic background, and image blocking histogram and so on. Through connected component labeling algorithm, we can get the number of moving objects accurately, the boundary information of image and detect targets easily. We study the detection of camouflage moving objects in the dynamic background, propose two camouflage target detection algorithms. The main points are as follows:1. Pixel labeled algorithm based on equivalence class method of connecting area in binary images. This article exploits special equivalence class to mark connecting area in binary images. According to the characteristics of the distribution on image pixels value scan the image, and then a temporary labeling image is produced. Meanwhile, connections of connected components are obtained in the equivalence table. By processing the equivalent of the mark, ascertaining the right distribution and count of connecting,a labeled image is produced. Experiments show that it can avoid the problem of repeating markers that most of the algorithms have to deal with, and exactly detect any connecting areas in binary images, which contains complicated shapes.2. Research on the gray value change of video which contains camouflaged moving objects. Some of the moving target is very similar to the color of the surrounding environment, which is called camouflage. In order to detect such camouflage targets, through experiments, obtain the gray value of camouflaged moving objects in video, and observe the changes of a particular pixel gray value in the continuous video. The analysis of gray value variation of the specific pixel lays a good foundation for researching on detection of camouflaged moving objects.3. The detection of camouflaged moving objects in gentle changed background. According to the characteristics of the gray value change of video which contains camouflaged moving objects. We put forward two methods for defecting camouflaged moving objects. One approach is based on the energy of Laws, this article redefined the calculation method on energy of Laws, calculating the pixel energy value of Laws in the current frame, from continuous Multi-frame image, according to the energy of Laws to detect target. The other is to analysis gray value changes in the current frame, according to gray value's varying frequency to detect target. Varying frequency in the text refers to the number of the signal alternating positive and negative relative to the local or instantaneous mean in a period of time. The signal representation pixel value changes with time. The experiment shows that, these two methods are effective for detecting camouflaged moving objects.4. Using Block histogram to detect target in dynamic background. In the complex background, such as shaking of the leaves, rippling water waves, etc, researchers use Gaussian mixture modeling to detect target usually. We have done some preliminary research in block histogram to solve the problems of computational complexity in Gaussian mixture modeling in this paper.At present, the algorithm is effective for detecting camouflaged moving objects only in slight change background, when the background changes greatly, (for example, the wind is very large, tree swing violently), further study on camouflage detection is required. Some primary works are done in soft realization. We only judge the general motion object region. Determining the exact location of the object is worth further studying.