Security Protection And Intrusion Alarm Techniques Using Image Processing And Support Vector Data Description

The devices using the traditional security protection and intrusion alarm techniques are simple, and their installationsites are usually visible. Sometimes intruders can easily evade the detectors, under the circumstances the security protection system can not take its own performance effectively. On the other hand, the traditional detectors detect the intruders passively and can not automatically analyze the condition of the protected area. With the development of the video monitoring system, the monitor cameras are widely used. When the video monitoring system operating, monitoring by mankind is not only high cost, but also usually failed to report some illegal intrusions. Contraposing to the problems above, new methods based on video monitoring images (VMIs) are proposed in this dissertation. These methods process and classify the VMIs to recognise abnormalities in the protected area and alarm will be reported automatically.When abnormalities happen, the VMIs are to change necessarily too. Contraposing to the characters of the VMIs, a new method of automatic intrusion alarm (AIA) based on difference image is proposed in this dissertation. The difference image is obtained by subtracting the normality's VMI from the real-time image. The two images are more similar, the largest gray value (LGV) of the difference image and the number of nonzero terms (NNT) in its histogram vector are smaller; the two images are greater different, the LGV of the difference image and the NNT in its histogram vector are higher. The change of these two parameters can reflect the differences of the VMIs. So they can be used as the indexes, if they are all greater than the presetting thresholds, it shows that illegal intrusion has happened in the protected area, then alarm will be reported automatically.In order to extract the main features of the image and reduce the input vector dimensions of classification, an image processing method based on the wavelet transform (WT) is studied in this dissertation. Decomposing an image with WT, four components as approximation component, horizontal component, vertical component and diagonal component are obtained. After decomposition each time, four components are obtained and each component's size is 1/4 of the component before the decomposition; the features of the original image are held in the approximation component, so the features information of the original image is condensed.When the important places are protected, illegal intrusions or other abnormalities are happenstances. The normalities's VMIs are easily obtained, in contrast, the abnormalities's VMIs are less and indeterminable to a great extent. Contraposing to the condition above, an AIA method adopting VMIs and support vector data description (SVDD) is proposed in this dissertation. The VMIs are decomposed with WT firstly, and then a one-class classifier can be constructed only using the normalities's VMIs. The experiment results show that the classifier can distinguish the normalities's VMIs and the abnormalites's VMIs effectively, so it can automatically recognise the condition of the protected area; by the approximation component as the input vector of the classifier, it ruduces the input vector dimensions greatly and improves the recognising and classifying efficiency, also it has a better recognising and classifying effect.