KR20090111207A - Mothed for classifying Surveillance object and Surveillance system thereof - Google Patents

Abstract

PURPOSE: A monitor distinguishing method and a monitoring system for detecting an object are provided to improve the performance of the monitoring system due to the quick search of the object. CONSTITUTION: A monitoring system for detecting an object is as follows. An input image about the monitoring region is input(S110). The detection image about the monitoring target area is detected the mobile object is detected with the monitoring object(S120). The detection image is converted to the Hilbert image(S130). The identification object is selected as an object of the target image.

Description

Surveillance object identification method and its monitoring system {Mothed for classifying Surveillance object and Surveillance system

The present invention relates to a method for identifying a surveillance object and a monitoring system thereof, and more particularly, to a surveillance object identification method for monitoring a specific area as a surveillance area and detecting a moving object within the surveillance area by installing a camera in a specific area. It relates to a surveillance system.

Surveillance system is a system that recognizes the target by processing the image inputted through a camera installed in a specific place or a mobile, and monitors a specific area. Such a surveillance system enables a small number of managers to efficiently control or manage various situations occurring at remote sites.

This surveillance system serves as a means to most effectively respond to various anti-social criminal acts in the modern society with a small number of persons and can be used as an information medium for judging the occurrence of various accidents. .

Human-centered surveillance systems can cause problems due to fatigue and reduced attention, and simple video equipment-based surveillance and security systems have limited coverage. Thus, the video surveillance system attracts attention.

In order to efficiently operate the video surveillance system, the video surveillance system can find out which object the detected object belongs to based on the shape and color information of the detected object. In other words, by transmitting the information determined about the detected object to the user, it is possible to provide a basis for configuring an intelligent surveillance system, not just transmitting the current image.

The present invention provides a monitoring object identification method and a monitoring system for establishing a representative model for a specific object such as a person or a car and comparing the detected object in the input image with the representative model to efficiently identify the object. The purpose.

The present invention, (a) receiving an input image; (b) detecting a moving object from the input image as a monitoring target, and detecting a detection image of the area of the monitoring target from the input image; And (c) converting the detected image into a Hilbert image in Hilbert space, comparing the Hilbert image with at least one representative image, and selecting an object of the representative image that is most similar as the identification object. To provide.

In step (a), the input image is input as a frame image of a predetermined time interval, and in step (b), two or more frame images are compared to extract a background image and a current image, and the background image and the current image. The monitored object can be detected from the difference of.

(C) step (c1) generating a preprocessed image by preprocessing the transformed image into a hilbert space, (c2) generating a hilbert image by mapping the preprocessed image to a hilbert space, (c3 Comparing the Hilbert image with the representative images, and (c4) selecting an object of an image determined to be the most similar to the Hilbert image among the representative images as the identification object.

The step (c1) may include converting the detected image into a binary image to obtain an object shape of the object to be monitored, extracting an outline of the object shape, and the pre-processing image may have a ratio of width and length to the outline. And maintaining and deforming the image including the outline to have a square shape to generate the preprocessed image.

In the step (c2), generating a Hilbert curve according to a preset Hilbert order, generating a split image obtained by dividing the preprocessed image into two dimensions according to the Hilbert order, and splitting the divided image along the Hilbert curve. And converting the HD into a one-dimensional Hilbert image.

The method may further include generating a representative image of the one-dimensional Hilbert image according to the Hilbert curve for each of the representative images according to the Hilbert order.

The step (c3) may include assigning different pixel values to each divided region of the Hilbert image and a divided region having an outline of each divided region of the representative image, wherein the pixel value and the representative of the Hilbert image Adding the pixel values of the image to obtain an added pixel value for each divided region, obtaining first distance values of the divided region having an outline near the added pixel value from the Hilbert image for the representative image; Obtaining second distance values of the divided region in which the outline near the added pixel value exists from the representative image from the representative image, and from the first distance values and the second distance values, The method may further include measuring similarity with respect to each of the representative images.

The similarity is measured by dividing the sum of the first distance values and the second distance values by the number of split images constituting the Hilbert image, and comparing the similarity between the Hilbert image and the representative image based on the similarity. Can be.

The object of the representative image having the smallest similarity with respect to the Hilbert image may be selected as the identification object, and may be identified as an unknown object when the similarity of the representative image with respect to the selected identification object is larger than a predetermined reference value.

Another aspect of the invention, the image input unit for receiving an input image; Detecting a moving object in the input image as a monitoring object, detecting a detection image of the area of the monitoring object in the input image, converting the detected image into a Hilbert image in Hilbert space, and converting the Hilbert image into at least one A controller configured to select an object of the representative image most similar to the representative images as the identification object; And a storage unit in which the representative images are stored.

The display unit may further include a display unit in which the detected image is separately displayed with respect to a background image in each of the frame images.

According to the monitoring object identification method and the monitoring system according to the present invention, it is possible to quickly detect and identify the moving object in the video surveillance, it is possible to improve the performance of the monitoring system.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a flowchart of a method S100 for identifying a monitored object, which is a preferred embodiment according to the present invention. In this case, the monitoring object identification method S100 may be implemented by the monitoring system 70 of FIG. 7. To this end, the monitoring object identification method S100 according to the present invention may be a program or algorithm such as software or firmware stored in the storage unit 73 of FIG. 7.

Referring to the drawings, the monitoring object identification method (S100) includes an input image input step (S110); Moving object detection step (S120); And object identification steps S130 to S170.

In the input image input step S110, an input image of a surveillance region to be monitored is received. In the moving object detecting step (S120), a moving object is detected as a monitoring object in the input image, and a detection image of a region of the monitoring object is detected from the input image.

In the monitoring object identification steps S130 to S170, the detected image is converted into a Hilbert image in the Hilbert space, and the object of the most representative representative image is selected as the identification object by comparing the Hilbert image with the plurality of representative images.

The method for identifying a monitored object (S100) sets representative models for a specific object such as a person or a car, compares the monitored object detected in the input image with the representative models, and selects the object most similar to the monitored object among the representative models. Identifies it as an identification object.

In the input image input step S110, the input image may be input as a frame image at a set time interval. In this case, the moving object detection step (S120) may compare two or more frame images, extract a background image and a current image, and detect the surveillance object from the difference between the background image and the current image.

In the moving object detecting step (S120), a moving object is detected as a monitoring object in the input image, and a detection image of a region of the monitoring object is detected from the input image. In this case, the detection image may be detected by an algorithm for extracting a background region from an input image, modeling a corresponding portion, and detecting a moving region through comparison with a modeled operation.

The monitoring object identification steps S130 to S170 may include a preprocessing step S130, a Hilbert image generation step S140 and S150, an image comparison step S160, and an identification object selection step S170.

In the preprocessing step (S130), the preprocessed image is generated by preprocessing the transformed image into the Hilbert space. In the Hilbert image generation steps (S140 and S150), the Hilbert image is generated by mapping the preprocessed image into the Hilbert space.

In the image comparison step (S160), the Hilbert image is compared with the representative images. In the identifying object selecting step (S170), the object of the image determined to be the most similar to the Hilbert image among the representative images is selected as the identifying object.

In the preprocessing step (S130), since the similarity between the detected detection image and the representative images cannot be directly measured, the detected image is converted into a binary image to facilitate similarity measurement, and the converted binary image is converted into an outline and extracted from the size change. You can do it.

FIG. 2 illustrates a preprocessing step S130 for converting to a Hilbert image in the monitoring object identification method S100 of FIG. 1.

Referring to the drawing, the preprocessing step S130 may include a binary image conversion step S131, an outline extraction step S132, and an image deformation step S133.

In the binary image conversion step S131, the detected image 21 is converted into a binary image 22 to obtain an object shape 22a to be monitored. In the outline extraction step (S132), an outline 23a of an object shape is extracted.

In the image deformation step S133, an image 23 including an object shape outline is converted to generate a square preprocessed image 24. In this case, the square pre-processing image 24 may include an object-shaped outline 23a while maintaining a horizontal and vertical ratio.

The binary image 22 may be generated by setting the portion corresponding to the background to '0' and the shape of the object to be monitored 22a to be '1'. In addition, the outline may be extracted by converting the detected image so that the outline portion becomes '1' and the background and the inside of the object become '0' in the outline extraction step (S132).

As such, by converting the preprocessed image 24 into a square shape, the preprocessed image 24 may be prepared to be mapped in the Hilbert space.

3A to 3C illustrate an image segmentation method according to the Hilbert order. 4A shows a two-dimensional segmented image 40a before mapping to Hilbert space. 4B shows a Hilbert image 40b mapped in one dimension in Hilbert space.

In the Hilbert image generation steps (S140 and S150), the Hilbert image is generated by mapping the preprocessed image into the Hilbert space. To this end, the Hilbert image generation steps S140 and S150 may include a Hilbert curve generation step, a split image generation step S140, and a Hilbert image conversion step S150.

In the Hilbert curve generation step, a Hilbert curve is generated according to a preset Hilbert order. In the split image generation step (S140), the split image 40a is generated by dividing the preprocessed image 24 in two dimensions according to the Hilbert order. In the Hilbert image conversion step (S150), the divided image 40a is converted into a one-dimensional Hilbert image 40b along the Hilbert curve 41.

In the Hilbert curve generation step, the Hilbert curve 41 is generated according to a preset Hilbert order, and the method of dividing the preprocessed image 24 and the Hilbert curve 41 vary according to the Hilbert order. The Hilbert order is a factor that determines the size of the two-dimensional image on which the Hilbert curve is generated. When the Hilbert order is d, the size of the Hilbert space is multiplied by the d power of 2 times the d power of 2.

A Hilbert curve can be generated using the set Hilbert order. The relationship between the Hilbert order and the size of the Hilbert curve is based on the relationship shown in FIGS. 3A to 3C. 3A to 3C are results of generating Hilbert curves when the orders are 1 to 3, respectively.

In this case, the amount of calculation increases as the Hilbert curve increases, but when the Hilbert curve is too small, the shape of the outline is not clear, and thus, the identification may occur.

In the divided image generation step (FIG. 4A), the divided image 40a is generated by dividing the preprocessed image 24 in two dimensions according to the Hilbert order. Each pixel value may be allocated to the divided region of the divided image 40a. have. In this case, a pixel value of '1' may be assigned to each divided region in which an outline exists, and a pixel value of '0' may be assigned to a divided region in which an outline does not exist.

In the embodiment shown in FIGS. 4A and 4B, a Hilbert curve 41 having a Hilbert order of 2 and a shape corresponding thereto is generated, and a preprocessed image is divided along the Hilbert curve 41 to generate a divided image 40a. . In addition, each division area is assigned a pixel value of 0 or 1 depending on the shape of the outline.

In the Hilbert image conversion step (FIG. 4B), the divided image 40a is converted into a one-dimensional Hilbert image 40b along the Hilbert curve 41, and pixel values assigned along the Hilbert curve 41 for each segmented area. By mapping, the one-dimensional Hilbert image 40b may be generated from the two-dimensional segmented image 40a. In this case, each segment may be a Hilbert pixel that is a comparison unit with representative images.

In this case, when measuring the similarity between the Hilbert image and the representative image, the similarity measurement time can be shortened by allowing the distance value to be measured in one dimension rather than in two-dimensional space.

In the image comparison step (S160), the Hilbert image is compared with the representative images. To this end, in the image comparison step (S160), the Hilbert image is compared with each representative image, and the distance value between Hilbert pixels for comparison between the two images is measured, and accordingly, between the Hilbert image and each representative image. Similar degrees can be calculated.

To this end, the method may further include generating a representative image of the one-dimensional Hilbert image according to the Hilbert curve for each representative image according to the Hilbert order. At this time. At least one or more preferably, a database for each representative image of the plurality of representative models may be generated and stored in advance. In this case, the representative image may be an image defined in the Hilbert space generated in one dimension according to a preset Hilbert curve.

5A to 5C illustrate embodiments of representative models for comparing the similarity with the Hilbert image. 5A is an embodiment that is a representative model for one person, and FIG. 5B is an embodiment of a representative model for a two person group. Each of (a), (b), and (c) of FIG. 5C is an example in which different angles, that is, 30 degrees, 60 degrees, and 80 degrees are rotated with respect to the representative model of the same vehicle, respectively.

FIG. 6 illustrates a method for measuring a similar degree between the image of the representative model and the Hilbert image by measuring the Hilbert distance value from the image of the representative model and the Hilbert image.

Referring to the drawings, the image comparison step S160 may include a Hilbert pixel value assignment step, a pixel value addition step, a first distance value calculation step, a second distance value calculation step, and a similarity measurement step.

In the Hilbert pixel value assignment step, different pixel values are assigned to Hilbert pixels in which the respective outlines of the Hilbert image and the representative image exist. That is, as in the embodiment illustrated in FIG. 6, '1' is assigned to the Hilbert pixel in which the outline of the Hilbert image 61 is arranged in one dimension along the Hilbert curve, and in one dimension along the Hilbert curve. '2' may be allocated to the Hilbert pixel in which the outline of the arranged representative image 62 exists.

In the pixel value adding step, the pixel value of the Hilbert image and the pixel value of the representative image may be summed to obtain an added pixel value for each Hilbert pixel. In this case, in the pixel value adding step, the sum of the pixel values of the Hilbert image and the pixel values of the representative image may be obtained, and the addition pixel value may be obtained for each Hilbert pixel of the addition image 63.

In the drawing, when an Hilbert pixel has an outline in each Hilbert image and the representative image, the corresponding Hilbert pixel of the addition image 63 is the pixel value '1' of the Hilbert image plus the pixel value '2' of the representative image, '3'. May be assigned.

In the first distance value calculating step, the first distance value 64 of the Hilbert pixel having an outline near the added pixel value from the Hilbert image may be obtained. In this case, the first distance value 64 may be the number of Hilbert pixels dropped from the Hilbert pixel having the pixel value of '1' to the nearest Hilbert pixel having the pixel value of '2' or '3' in the added image. .

In the second distance value calculating step, a second distance value of the Hilbert pixel having an outline near the added pixel value with respect to the Hilbert image may be obtained from the representative image. In this case, the second distance value may be the number of Hilbert pixels dropped from the Hilbert pixel having the pixel value of '3' to the nearest Hilbert pixel having the pixel value of '1' or '3' in the added image.

In this case, when the added pixel value is '3', since the outline exists at the Hilbert pixel at the same position, the first distance value and the second distance value may be '0'.

In the similarity measuring step, the similarity degree of each representative image of the Hilbert image may be measured from the first distance values and the second distance values. In this case, the similarity may be measured by dividing the sum of the first distance values and the second distance values by the number of split images constituting the Hilbert image, that is, the number of Hilbert pixels.

In this case, the degree of similarity of the Hilbert image to each representative image may be compared based on the similarity of the representative images of the Hilbert image with respect to the input image.

In the identifying object selecting step (S170), the object of the image determined to be the most similar to the Hilbert image among the representative images is selected as the identifying object. In this case, the object of the representative image having the smallest similarity with respect to the Hilbert image may be selected as the identification object.

Meanwhile, when the similarity of the representative image with respect to the selected identification object is larger than the set reference value, it may be identified as an unknown object.

According to the present invention, it is possible to quickly detect and identify moving objects in video surveillance, thereby improving the performance of the surveillance system.

The monitoring object identification method S100 according to the present invention may reduce the time for detecting the moving object and identifying the detected object in the video surveillance, thereby enabling detection and identification in real time.

Accordingly, the performance of the monitoring system can be improved, and the user's satisfaction with the products of the monitoring system can be improved. In addition, fast performance and low memory requirements can be designed to be easily applied to small embedded devices through miniaturization and modularity.

7 shows a schematic block diagram of another preferred embodiment of the monitoring system 70 according to the present invention.

Referring to the drawings, the monitoring system 70 may be controlled by the monitoring object identification method (S100) shown in FIG. Therefore, the same matters as in the monitoring object identification method S100 illustrated in FIG. 1 are referred to this, and detailed description may be omitted.

Surveillance system 70 includes an image input unit 71; Control unit 72; Storage unit 73; And a display unit 74.

The image input unit 71 receives an input image of a surveillance area to be monitored. The controller 72 detects a moving object in the input image as a monitoring object, detects a detection image of a region to be monitored in the input image, converts the detected image into a Hilbert image in the Hilbert space, and converts the Hilbert image into a plurality of representative images. The object of the most representative representative image is selected as the identifying object by comparing with the images.

The storage unit 73 stores representative images of representative objects to be compared in order to identify a monitoring object that is a moving object included in the input image. The display unit 74 may display a detection area, which is an area for the monitoring target, in the input image separately from the background image.

The input image is input as a frame image of a set time interval, and compares two or more frame images, extracts a background image and a current image, detects a monitoring target from the difference between the background image and the current image, and monitors the input image in the input image. An image of the area of the object to be detected may be detected as a surveillance image.

In addition, in order to convert the detected image into a Hilbert space, a preprocessed image is generated, a preprocessed image is mapped to a Hilbert space, a Hilbert image is generated, a Hilbert image is compared with representative images, and among the representative images, The object of the image determined to be the most similar may be selected as the identification object.

Convert the detected image into a binary image to obtain the object shape of the object to be monitored, extract the outline of the object shape, and include the preprocessed image while maintaining the ratio of the horizontal and vertical lines, and include the image including the outline in a square shape. The image may be modified to generate a preprocessed image.

A Hilbert curve may be generated according to a preset Hilbert order, a pre-processed image may be generated by dividing the preprocessed image into two-dimensionally divided Hilbert orders, and the divided image may be converted into a one-dimensional Hilbert image along the Hilbert curve.

The storage unit 73 may store the representative image in the form of a database. The database for the representative image may be generated and stored as one-dimensional Hilbert image according to the Hilbert curve for each representative image according to the Hilbert order.

In the display unit 74, the detection area of the input image is divided with respect to the background image and displayed by a separate display such as a quadrangle, so that the user can easily monitor the moving object.

According to the present invention, it is possible to quickly detect and identify moving objects in video surveillance, thereby improving the performance of the surveillance system.

Surveillance system 70 according to the present invention, by reducing the time to detect the moving object and the detected object in the video surveillance, it is possible to make detection and identification in real time.

Accordingly, the performance of the monitoring system can be improved, and the user's satisfaction with the products of the monitoring system can be improved. In addition, fast performance and low memory requirements can be designed to be easily applied to small embedded devices through miniaturization and modularity.

Although the present invention has been described with reference to the embodiments shown in the drawings, these are merely exemplary, and those skilled in the art will understand that various modifications and equivalent other embodiments are possible. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

1 is a flowchart schematically illustrating a method for identifying a monitoring object as a preferred embodiment according to the present invention.

FIG. 2 is a diagram illustrating a Hilbert image preprocessing step in the monitoring object identification method of FIG. 1.

3A to 3C are diagrams illustrating an image segmentation method according to a Hilbert order.

4A is a diagram illustrating a two-dimensional segmented image before mapping to Hilbert space.

4B is a diagram illustrating a Hilbert image mapped in one dimension in Hilbert space.

5A to 5C are diagrams illustrating embodiments of representative models for comparing a similarity with a Hilbert image.

6 is a diagram illustrating a method of measuring a similar degree between an image of a representative model and a Hilbert image by measuring a Hilbert distance value from an image of the representative model and a Hilbert image.

7 is a block diagram schematically showing a monitoring system as another preferred embodiment according to the present invention.

Claims (19)

(a) receiving an input image; (b) detecting a moving object from the input image as a monitoring target, and detecting a detection image of the area of the monitoring target from the input image; And (c) converting the detected image into a Hilbert image in Hilbert space, and comparing the Hilbert image with at least one representative image to select an object of the most representative representative image as an identification object. The method of claim 1, In the step (a), the input image is input as a frame image of a set time interval, And (b) comparing two or more frame images, extracting a background image and a current image, and detecting the surveillance object from a difference between the background image and the current image. The method of claim 1, In step (c), (c1) generating a preprocessed image by preprocessing the transformed image to convert to the Hilbert space; (c2) generating the Hilbert image by mapping the preprocessed image to the Hilbert space; (c3) comparing the Hilbert image with the representative images, and and (c4) selecting an object of an image determined to be the most similar to the Hilbert image among the representative images as an identification object. The method of claim 3, In step (c1), Converting the detected image into a binary image to obtain a shape of an object to be monitored; Extracting the outline of the object shape, and And the preprocessed image including the outline while maintaining a ratio of width and length, and generating the preprocessed image by modifying the image including the outline to have a square shape. The method of claim 4, wherein And extracting the outline by converting the detected image such that the outline portion is '1' and the background and the inside of the object are '0'. The method of claim 4, wherein In step (c2), Generating a Hilbert curve by a preset Hilbert order, Generating a divided image by dividing the preprocessed image into two dimensions according to the Hilbert order; And converting the split image into a one-dimensional Hilbert image along the Hilbert curve. The method of claim 6, And a pixel value of '1' is assigned to each divided region of the divided image in which the outline exists, and a pixel value of '0' is assigned to the divided region in which the outline does not exist. The method of claim 6, And mapping the pixel values along the Hilbert curve to each of the divided regions to generate the Hilbert image of the two-dimensional segmented image. The method of claim 8, And generating a representative image of the one-dimensional Hilbert image according to the Hilbert curve for each of the representative images according to the Hilbert order. The method of claim 9, In step (c3), Allocating different pixel values to each divided region of the Hilbert image and a divided region having an outline of each divided region of the representative image; Obtaining an added pixel value for each divided region by adding the pixel value of the Hilbert image and the pixel value of the representative image; Obtaining first distance values of the divided region where an outline close to the representative pixel value exists from the Hilbert image with respect to the representative image; Obtaining second distance values of the divided region in which an outline close to the Hilbert image exists from the representative image from the representative image; and And measuring similarity of the representative images of each of the Hilbert images from the first distance values and the second distance values. The method of claim 10, The similarity is measured by dividing the sum of the first distance values and the second distance values by the number of split images constituting the Hilbert image, And a degree of similarity between the Hilbert image and the representative image based on the similarity. The method of claim 11, The object of the representative image having the smallest similarity with respect to the Hilbert image is selected as the identification object, And identifying the unknown object when the similarity of the representative image with respect to the selected identification object is larger than a set reference value. An image input unit which receives an input image; Detecting a moving object in the input image as a monitoring object, detecting a detection image of the area of the monitoring object in the input image, converting the detected image into a Hilbert image in Hilbert space, and converting the Hilbert image into at least one A controller configured to select an object of the representative image most similar to the representative images as the identification object; And Surveillance system having a storage unit for storing the representative image. The method of claim 13, The monitoring system for inputting the input image as a frame image of a predetermined time interval, comparing two or more of the frame image, extracts a background image and the current image, and detects the monitoring target from the difference between the background image and the current image . The method of claim 13, In order to convert the detected image into a Hilbert space, a preprocessed image is generated, a preprocessed image is mapped into a Hilbert space, a Hilbert image is generated, the Hilbert image is compared with the representative images, and among the representative images. And selecting an object of an image determined to be the most similar to the Hilbert image as an identification object. The method of claim 15, Converting the detected image into a binary image to obtain an object shape of the object to be monitored, extracting the outline of the object shape, and including the preprocessed image while maintaining the horizontal and vertical ratios of the outline, and including the outline Surveillance system for generating the pre-processing image by transforming the image to a square shape. The method of claim 13, Monitoring to generate a Hilbert curve according to a preset Hilbert order, to generate a split image divided into two dimensions according to the Hilbert order, and to convert the split image into a one-dimensional Hilbert image along the Hilbert curve. system. The method of claim 17, And the representative image is generated as a one-dimensional Hilbert image according to the Hilbert curve for each of the representative images according to the Hilbert order. The method of claim 14, And a display unit which displays the detected image separately from the background image in each of the frame images.

Images