Security System Design Based on Intelligent Video Surveillance (2)
Relevant information:
Security System Design Based on Intelligent Video Surveillance (1)
3 Implementation of intelligent video analysis
The acquisition and analysis of video images is mainly done by an embedded microprocessor built into the front camera. This method of data processing allows the system to analyze the original or closest original image and make quick and accurate judgments in the first place.
A complete video image analysis process requires a combination of image processing technology, pattern recognition technology and other technical means to achieve better practical results. The working process includes image preprocessing, image segmentation, feature extraction and image classification. The work flow chart is shown in Figure 3.
The image recognition design of the system is realized by the idea of ​​motion detection: firstly, the background is restored according to the statistical value of the pixel values ​​of each coordinate in the whole sequence, and if there is an abnormal situation, it is extracted; then the statistical method is used to identify the abnormal situation. The category of the category.
The image recognition is mainly realized by the inter-frame change detection method, and the basic flow is divided into:
(1) Image preprocessing, using various special techniques to highlight certain details in the image and weaken or eliminate irrelevant information according to the blurring of the image, thereby achieving the purpose of enhancing the overall or local features of the image.
(2) Background restoration and anomaly extraction of the image, recovering the background of the image according to the statistical information of the pixel values ​​in the entire sequence, and then subtracting the background from the recovered background to extract the region where the abnormality has occurred. (3) Image classification, using the current frame and the recovered static background to subtract, extracting all areas where an abnormal situation may occur.
3.1 Image Preprocessing
The common image enhancement can be divided into two categories: spatial domain and frequency domain. The spatial domain enhancement is directly processed in the space where the image is located, and the pixel gray value of the image is directly processed. Spatial domain image enhancement techniques can be described using Equation 1:
Where: F(x, y) is the image before processing, G(x, y) is the processed image, and H(x, y) is the spatial operation function. The frequency domain image enhancement is to image the original space to some kind. The form is converted to another space, then processed using the unique properties of the transform space, and finally converted back into the original space. The process can be described in Figure 4:
3.2 Image background recovery and abnormal extraction
There is a strong correlation between the frame and the frame of the video sequence. If only single frame information is used for analysis and processing, the error rate is high, and the current analysis method is better in combination with multi-frame processing. Based on this idea, the background can be restored based on the statistical information of the pixel values ​​at the respective coordinates in the entire sequence. The system designed in this paper uses static background recovery for processing:
First, we define the image sequence as B(x, y, i), where x, y represent the spatial coordinates, i represents the number of frames (i = 1, ..., N), and N is the total number of frames in the sequence. The video frame difference CDM reflects the grayscale variation between adjacent frames:
Where: Threshold T is used to remove noise. For a fixed coordinate position (x, y), CDM(x, y, i) can be expressed as a function of the number of frames i, which records the curve of the pixel point (x, y) along the time axis. This curve can be segmented according to whether CDM(x, y, i) is greater than zero and used as a set 
Said.
The system operation steps are as follows:
(1) reading the adjacent two frames of data, comparing them, and calculating their difference; (2) binarizing the obtained two frames of image differences, and binarizing the image in a specific corrosion window Corrosion treatment is performed under conditions, and then the standard interframe shift of the image after the etching treatment is calculated; (3) steps (1) and (2) are repeated; (4) the maximum length of tracking for each pixel is set to 0, and record the middle frame label in its maximum length; (5) traverse the entire sequence, track and record the maximum length of a single point continuously 0; (6) traverse the entire frame image, set the background data, and recover the static background.
3.3 Image classification
After extracting the target object from a complex scene, in order to facilitate the identification, it is necessary to measure and calculate the size feature and shape feature of the target object. These features must have certain stability with respect to a particular object. For example, when the image is rotated or translated, the area and circumference of the object do not change significantly; when the image is ingested due to the difference in distance between the target object and the camera When the image size is different, the scale feature does not change. The system can use these more stable features to distinguish different objects to accurately identify the target entering the scene. In this system, the area and scale features are mainly used to distinguish moving objects.
Image classification system uses statistical pattern recognition method which several predefined type or category, which the image that may contain one or more objects, and each object belonging to a certain category or type previously defined class. Completed wherein The devices that "classify" work are "classifiers". There are many types of classifiers, such as parametric and non-parametric, linear and non-linear. This system design selects a linear classifier and uses the minimum distance classification method to identify and classify objects. The method uses the distance between the points of the input pattern and the feature space as a template as a criterion for classification. There are m categories of images with categories W1, W2, ..., Wm. Now to determine which of the m categories a given image is, you can judge by extracting the features of the image. .
When there are many image categories, the features are generally more. For the convenience of analysis and classification, it is possible to represent the feature vector X in the d-dimensional space by the d of the image (assuming that the image has d features), and if there are m categories, there are m such feature vectors.
Therefore, after extracting all the features from the image, the d-dimensional vector is composed, and the minimum distance classification method is used to match the feature vector of the image class in the sample library, if it is related to the i-th (0)
4 small knots
Aiming at the development status and existing problems in the field of video surveillance, based on the previous research results and the author's work practice over the years, this paper proposes a monitoring system design based on intelligent video analysis, which has been done from hardware architecture and software architecture. The analysis shows that a new method and idea for the design of the monitoring system is provided.
The application of intelligent video surveillance technology is very promising. The security management of parking lots, highways, factories and military bases is in urgent need of help, and with the continuous improvement of hardware processing capabilities and software analysis capabilities, the performance of intelligent video surveillance systems It will continue to improve, and new functions will continue to emerge. It will replace the existing traditional monitoring and open a new chapter in security management.
Steel Grating Fences,Steel Grating,Steel Fence
Steel Grating,Miscellaneous Co., Ltd. , http://www.steel-grating-manufacturer.com