Statistical Model Of Video Sequences And Target Visibility Analysis And Applications

Video analysis and processing are widely used in many fields such as commercial, industrial, military fields and so on. Detection and segmentation are important techniques in video analysis and processing because we desire to detect and extract the related video objects (VOB), we are interested in. For example, the applications are content-based video data indexing and querying, site monitoring system, infrared targets tracking and recognition system.Random experiment model is established for general video sequences in this paper. As multidimensional random signal, video sequence has such descriptions as sample set, random variable set and gray scale random variable. Other descriptions of random video sequence are introduced in terms of pixel algebra and video object. For experiment convenience, we describe the video sequence in this paper as gray scale random variable, and explore fully its statistical characteristics. With analysis and PC simulation, we have got the conclusions as following: video sequence is statistically dependent and has Marcov statistical characteristics in local spatial-temporal scope, differential video background sequence is spatial-temporal stationary and the residual noise on it is like white gaussian, and so on. These characteristics and descriptions are the basis or hypothesis of many video analysis and processing methods. This paper has done a lot of necessary research in this special field.Another important contribution of this paper is the research about visual ability. By analyzing several elements that have effects on visual ability, the contrast in neighborhood is given and the relation with the visual ability is discussed and simulated. From the research of error characteristic of detection, the signal to noise ratio in neighborhood and its relation with detection ability are discussed. Then the visual ability model and classification methods are established. On the basis of the models and experiments mentioned, three techniques for applications are discussed in this paper as noise characteristic parameter estimation, background estimation, TBD detection. The first two are simulated, analyzed and compared. In the noise characteristic parameter estimation, we pay much attention to how to choose its original parameter and analyze the influence of iterationfunction to detection result. In the background estimation, we mainly analyze relation between the size of the block and target, estimation time, and so on. The simulation platform of software designed in this paper is complete and can be widely used in the teaching and research.