With the increasingly complex social security condition, public security issues continue to deteriorate. To be more specific, the spiral of terrorism becomes never-ending and the urban security malady keeps worsening. In order to solve the problem, it is necessary to apply the visualized video surveillance system based on image processing.
The traditional video surveillance system relies on manual intervention and is tremendously labor intensive. It calls for concentration, vigilance and response ability to abnormal situations, which makes manual surveillance a very challenging task, especially in the scene with dense moving objects. Therefore, the real-time performance of traditional visual monitoring system is significantly poor.
In order to address the problem of social security well, the visualized monitoring system based on image processing has been optimized compared to traditional one. The visualized surveillance system can not only analyze ,automatically, the image sequence captured by the camera but also locate, recognize and track the target in different scenes. If the abnormal behavior is detected, it will trigger the automatic procession, such as alarm, automatic video recording, etc, which significantly improves the real-time performance, and ,at the same time, reduces the human resource consumption and raises the reliability.
In visualized monitoring system, target detection is one of the major applications of image processing. The purpose of target detection is to extract the moving area from the background images. This technology is based on background modeling and precise foreground segmentation, which separates foreground objects from background as well as locates the objects.
In target detection, frame image segmentation is a main step. Without considering the information of the frames around, it can be regarded as the segmentation of still images. Common image segmentation techniques can be roughly divided into edge-based methods and region-based methods.
-
Threshold Segmentation Method
Threshold method is widely applied in gray image segmentation, which first determines a threshold within the range of gray-scale value of the image, then compares the gray value of each pixel with the threshold, and sequentially, divides the corresponding pixels into two categories according to the result of comparison, namely the larger one as well as the lower one. These two kinds of pixels belong to ,respectively, two kinds of regions (foreground and background) in the image. Thus, regions are separated successfully by classifying the pixels according to the threshold. -
Edge Detection Method
Applying edge detection method, the areas are separated by detecting the boundary of different regions. Edges tend to be presented in the form of intensity mutation, which can be defined as discontinuity of local features, such as gray-scale value mutation, texture structure mutation and so on. Indicating the end of one region and the beginning of another, the edges are likely to imply the determining information about the shape of an object, which not only sharply reduces the amount of information to be processed, but also remains the boundary structures of the objects. Therefore, edge detection matters a lot when it comes to more complicated situations. -
Region Tracking Method
The two methods above both focus on the difference of pixels, while region tracking is to seek similar groups of pixels, which correspond to the planes or objects of the physical world. Starting from one pixel, it is based on the principle of consistency (this consistency can be gray-scale value, color, gradient or other characteristics) to gradually increase the pixels, i.e. region growth. For the region composed of these pixels, a uniform measure function is applied to test its uniformity. If it is true, the region will continue to expand until the result is false. There are two common methods, namely, region growing method as well as region splitting and merging method.
The former one starts from the point that satisfies the detection requirement and “grow” objects in all directions. Due to the fact that the difference of gray-scale value or color in the same area is extremely small, I is assumed to be a qualified object , then the property of all its adjacent points can be checked. If the detection criteria are satisfied, the adjacent points will be merged into the original block. When the adjacent points “grow” into a new I, the above process will repeat ,with the new I, until there is no acceptable adjacent point.
The region splitting and merging method divides the image into initial regions before merging these areas. Thus, the performance of segmentation is enhanced progressively until the image is finally separated into the tiniest uniform regions.
In addition to target detection, the visualized surveillance system also classifies and tracks the targets in dynamic scenes. On such basis, it is able to analyze the behavior of the targets, so that it can not only fulfill the daily tasks, but also respond to the abnormal situations.
In conclusion, the visualized surveillance system based on image processing meets the demand of modern society, and is capable of addressing social security issues better.