KR20060100341A - Image monitoring system for object identification - Google Patents

Abstract

The recording method used in the current security system records the entire image captured by the camera without image processing.

In order to identify the cause after an accident or an incident, all the recorded information should be searched and the corresponding video should be searched. Of course, if you know the accident time and the scene, the search time will be shortened. If not, you will have to spend a lot of time and effort to find the recorded video.

In order to reduce the size of the video file recorded by the current security management system, the video recording method is shortening the video recording storage time by using the video recording method by detecting the motion or the recording method when the external sensor operates. .

The present invention aims to provide an auxiliary means for quickly searching for a desired image from recorded video information to quickly identify the cause of an accident.

Most of the security-related video search is more than 95% people search. Therefore, by identifying the object called human from the real-time image being captured, only the human image is extracted and stored as a still image, and later, when an accident occurs, the human image is searched first to determine the occurrence time and search for the corresponding video. It is an invention that can help to deal with the case by making it clear.

In addition, to extract a feature from the extracted human image to compare the still images recorded by a plurality of cameras to search for a moving object having the same characteristics to provide a function that can track the path of a specific person.

In order to implement the present invention, first, the extraction algorithm that can identify a person, the algorithm that extracts a feature that can distinguish the extracted person from other people's images, and the algorithm that extracts similar images by comparing the extracted characteristics are combined. It is possible to provide a system for retrieving an image that can solve the problem.

Function to track the object by linking the camera with the path tracking and pan-tilt using the identified person object information, alarming function when entering the area with specific threshold value, direction detection function In addition, it can provide an infrastructure to build a surveillance system that can be applied, such as the presence of object accumulation.

Security video, video search, object management, surveillance system

Description

 Image Monitoring System for Object Identification

Figure 1. Process Flow Diagram for Identifying People in Motion Images of the Invention

Figure 2. Exemplary drawing method of creating a background image of the human object extraction step of the present invention

Figure 3. Exemplary image of converting RGB image to HSI image and labeling the converted saturation image during human object extraction step of the present invention

Figure 4. Exemplary diagram showing a process for extracting a human object of the present invention

Figure 5. Process Flow Diagram for Searching Stored Images of the Invention

Figure 6. Exemplary folder stored during the image search step of the present invention

Figure 7. Exemplary object image stored during the image search step of the present invention

Figure 8. Illustrated search result of the image search step of the present invention

Figure 9. Exemplary diagram showing the search results collected by playing the video during the image search step of the present invention

Figure 10. List of Object Management Variables of the Present Invention

Figure 11. Object Management Database Table Schema of the Present Invention

The recording method used in the current security system records the entire image taken by the camera without image processing.

In order to identify the cause after an accident or an incident, all the recorded information should be searched and the corresponding video should be searched. Of course, if you know the accident time and the scene, the search time is shortened, but if not, you can find the recorded accident video by taking a lot of time and effort.

In the current security management system, a video recording method uses a method of detecting a motion or recording a video by an external sensor to shorten the recording time of the video recording.

Most of the security-related video search is more than 95% people search. Therefore, by identifying the object called human from the image being captured, it extracts only the human image and keeps it as a still image.In the event of an accident later, the human image is searched first to find out the recording time, and automatically searches for the corresponding video. It is an invention that can help to deal with the case by identifying the cause.

In order to shorten the search time, the feature is extracted from the extracted human image and compares the still images recorded in a plurality of cameras to provide a function to track the path of a specific person by searching for still images with similar characteristics.

In addition, the function can track the object by linking the camera with the path tracking and pan-tilt using the identified person's object information, alarming function when entering the area with a certain threshold value, and direction detection. It provides an infrastructure to build a surveillance system that can be applied for functions, presence of objects, etc.

The most important factor in the present invention is an algorithm for extracting image information for identifying a person from image information captured in real time.

1 is an algorithm for extracting a person by detecting a motion in a captured image, and in step 101 of extracting a background screen to extract a person, the motion is extracted by a difference between the background screen and a real-time image. The accuracy of the extracted object depends on how accurate the background image is.

Therefore, the algorithm for extracting the background is more important in the outdoors than in the interior. The identification of differences in day and night brightness should be prevented.

In the present invention, a background image is created using two methods. First, the background image is stored at regular intervals. At this time, the image is acquired when there is no movement. Secondly, if it is determined that there is no motion by comparing the acquired image with the previous image, the background image is managed as the average value of the sum of the RGB values of the current background image and the two images.

2 illustrates an algorithm for extracting a background screen. The initial background image is input by accumulating a certain number of frames manually and accumulating the average.

When the background image is generated, the subtraction operation is performed to extract the motion image after the subtraction operation with the later image. If there is no motion image, the motion image is accumulated on the current background image to obtain a more real-time background image. do. Alternatively, if it is determined that there is no movement after a predetermined time, the corresponding image is accumulated in the background image (103).

The background image thus prepared and the currently captured image are subtracted to detect motion.

The size of the extracted motion image is compared (104). At this time, if the size of the extracted motion image is larger than a certain size, it is likely that there are two or more objects or completely different objects. Therefore, in this case, only the image is kept and object management is not performed (108).

When the extracted motion image is one person, the extracted motion image is converted into an RGB value into an HSI value (105). An image having a saturation value is created (106) by applying hybrid median filtering suitable for boundary preservation, and a labeling operation is performed (107). Hair colors vary widely from country to country to define the hair color for a particular country. Koreans usually have black hair. Therefore, when labeling work consisting of saturation value, more than 95% of the head value is distinguished as 250 or more among 0 ~ 255 values.

FIG. 3 shows three images of a person (object) extracted as a motion image, an image obtained by HSI conversion of the image, and a saturation image.

The reason why the labeling work is applied to the saturation image is that the saturation image shows the shape of the head more clearly than the hue image. You can see that it has a shape, so it is much easier to extract hair.

4 is a subtraction operation of a captured image 402 from a background image 401 to extract a motion image (403), HSI conversion (404), and then sequentially label the content (405) of a saturation image. This is the picture shown.

If it is determined that the size of the extracted motion image is one person, the position of the head is judged to determine whether the person is present. If it is determined that the person is given a unique ID to manage a variety of management variables (108). The reason is that the moving object in one camera must be managed continuously, so the extracted object must be assigned a unique ID to be managed so that the object can be tracked.

 At this time, the memory variables for managing the extracted human objects include the unique ID shown in FIG. 10, the created time, the expired time (the time disappeared from the screen), the average value of the color values of the upper body and the lower body, and the imaging position of the camera. , To prepare and manage the immediately moving direction, the photographed camera number, the photographed controller number, and the size of the acquired image. Images are stored in folders by location, camera, and date and time (Fig. 6). This is for the convenience of the search later. The variable information and the corresponding motion image path information that are managed when they are destroyed (when they disappear from the screen) are stored in the database table of FIG. 11.

 At this time, the unique identifier (ID) needs to have a unique file name in order to be useful when searching. To do this, it is necessary to put an image pickup position, a camera number, a date and time, an object number (number given at the moment), etc. in a unique identifier (ID).

 Next, a method of searching for and displaying information of a person who is similar to the person to be searched using the stored information will be described.

In the image retrieval algorithm of FIG. 5, the first thing to do is to select an image to be searched for and read information on the selected image (501). This can be selected from images stored for each camera installed at each position, as shown in FIG. 6.

In FIG. 6, the main reason for image search is to find people who have moved in a specific location when an event or accident occurs at a specific location. Therefore, as the search condition, the location information should be made higher (601). However, since the position-specific camera number 602 and the date 603 do not need to be prioritized, the camera number does not need to be prioritized over the date as shown in FIG. 6.

Searching for the folder in FIG. 6 shows all the images stored on the selected date as shown in FIG. 7. If an image to be searched is selected, the corresponding image is HSI-converted and the resulting Hue image is set as a reference value (502). The Hue image of the reference value is separated into upper and lower bodies and the value is compared with the UpperHue and LowerHue fields stored in the database.

Since the Hue values of all frames are not the same, a threshold value is set to search for all images having a difference within the error limits.

 The retrieved images are displayed together with the information of each image as shown in FIG. 8. 8 shows only the same images, but there are chances that other images may appear when the color information is similar by chance. Therefore, in this image retrieval algorithm, only the necessary images can be selected and viewed, and only the desired images can be reproduced and confirmed. As a result, as shown in FIG. 9, an image at a desired position and time zone can be reproduced and viewed. The timestamp of the video information recorded in MPEG4 or MJPEG associated with the generation time information, the controller information, and the camera information of the identified object information may be searched for information on a location where the selected object information is recorded.

The overall composition of the image search algorithm is as follows. The image search algorithm finds the desired location and time zone in the saved folder (Fig. 6), selects the desired object (People) in that time zone (Fig. 7), and searches the most similar objects within the margin of error with the selected object (People). FIG. 8 collects only desired images and plays them as a video (FIG. 9).

 In order to identify the cause after the accident, all recorded information is searched and the corresponding video is searched. Of course, if you know the accident time and the scene, the search time will be shortened. If not, you will have to spend a lot of time and effort to find the recorded video. Most of the cases have involved people. Therefore, if the user searches the recorded contents mainly for the video information to be searched, the search time can be greatly reduced. The present invention manages human-oriented information in a database during a movement to search for various image information to quickly identify the causes of incidents and accidents, thereby providing an efficient security management task. In addition, the function can track the object by linking the camera with the path tracking and pan-tilt using the identified person's object information, the alarm function when the trespassing area is assigned to a certain threshold value, and the direction. It can provide an infrastructure to build a surveillance system that can be applied for sensing functions and the presence of physical objects.

Claims (3)

 An image processing surveillance system, comprising: creating a background image for extracting a human object; Extracting a motion image by subtracting an image captured in real time and a background image created in the previous step; Determining whether the extracted motion image is a single person; Converting an RGB image to an HSI image; Creating an image to which a hybrid median filter is applied to an image having a saturation value; Identifying the head in the position and size by identifying the head in the image created in the step; When the extracted motion image is identified as one person, a unique identifier, a captured image surveillance controller name, a camera number, a generated time, an expired time, an upper body color value of the identified object, and a lower body color of the identified object Managing (Hue) values, an X-axis position of the camera of the identified object, a Y-axis position of the camera of the identified object, and a moving direction of the identified object; And storing the identified variable information in a database when the identified object is deleted from the camera.  An image processing monitoring system, comprising: converting a selected moving image into upper body color (Hue) and lower body color (Hue) information when a motion image is selected and a similar object is searched for in FIG. 7; By searching the information stored in the database with the extracted upper and lower body color (Hue) value to display the image information similar to the selected motion image as shown in Figure 8, the controller name, camera number, image capturing time information and the extracted image step; Video surveillance system having a feature comprising a.  In the image processing monitoring system, when the stored motion image is selected as shown in [Fig. 7] and [Fig. 8], it is recorded by the corresponding MPEG4 or MJPEG with reference to the generation time information, controller name and camera number stored in the database. Video surveillance system that can search the time stamp of the video and automatically search the video location where the selected motion video is recorded.

Images