Intrusion Detection Under Complex Scene

With the development of computer vision, artificial intelligence, pattern recognition technology, intelligent video analysis based on the perimeter intrusion detection has become the focus of the study. Compared with traditional intrusion detection sensors, video motion analysis-based intrusion detection technology has great advantages such as greater detection range, higher detection rate (POD) and lower false alarm rate (FAR). It also can hidden detect, remote detect and so on. So one can use it to replace various types of joint or line (or point) sensor for intrusion detection and alarm. In this paper, we study existing algorithms and analysis very step of them, then construct a set of intrusion detection algorithm based on video motion Analysis. Our algorithm improves efficiency.1 This paper provides the method to solve the problem caused by light mutations, which Gaussian mixture (MOG) model cannot handle with. It identifies the light mutation by using the number of the connected components and information of area ratio, in the meanwhile, establishes the Smooth Gaussian Model that combines sampling every two frames with IRR Moving Average Model, which meets the real-time needs and adapts light mutations fast as well, so that this method can reduce false positive.2 In the process of detecting shadows and denoising, considering the real-time and high efficiency, we use the method of shadow detection based on the normalized RGB. We also provides an efficient reconstruction algorithm that recoveries the prospects of real movement which combines binary image of integrated prospect with set operation, in order to solve the problem of shadow-removing by using color features. Besides, we propose a new method to maintain the integrity of target prospect with combination of the median filter and projection filter.3 In the process of targeting and tracking, we obtain a shape-based classification and a color-based particle filter. We also put forward the method of the adaptive tracking window based on the spatial information exchange. This method has improved the tracking accuracy and robustness.4 Based on the work above, we presents a set of intrusion detection algorithm under complex scenes, including fixing the moving target, filtering and classifying target, applying effectively to the switched scenes, judging target invasion, alarming and tracking. This method realizes three ways for identifying invasions (into, out, mix line). The experiments of intrusion detection have been carried out under complex scenes, which verify that the proposed techniques are effective and robust.The proposed algorithm performs better on the aspects of real-time, adaption, and interference immunity, which can meet the needs of high detection and low false alarm.