KR101294498B1 - Method of objects recognition for tunnel accident monitoring system - Google Patents

Abstract

The present invention relates to an object recognition method of a tunnel note system, and more particularly, to an object recognition method of a tunnel note system capable of classifying a type of a moving object according to a range of size ratios of detected objects.
In the object recognition method of a tunnel note system for detecting a moving object through a camera image, the object recognition method of the tunnel note system according to the present invention comprises the steps of: labeling a rectangle along the outer angle of the detected moving object in the image; Through the step of classifying the type of the moving object according to the range of the aspect ratio of the, it is possible to distinguish the type of the object it is possible to respond quickly in the tunnel situation.
In addition, by comparing the labeling ratio of the current frame image with the labeling ratio of the previous frame image, if the two ratios are different from each other, the overlapping of the moving objects can be detected by determining that the overlapping of the moving objects has occurred, and thus continuous tracking can be performed. have.

Description

{Method of objects recognition for tunnel accident monitoring system}

The present invention relates to an object recognition method of a tunnel note system, and more particularly, to an object recognition method of a tunnel note system capable of classifying a type of a moving object according to a range of size ratios of detected objects.

In general, vehicle accidents in confined spaces such as tunnels are more likely to cause serious accidents than on roads. In-tunnel image detection system is introduced for early response and management.

The tunnel image detection system detects types of accidents such as stationary vehicles, driving vehicles, pedestrians, and falling objects by analyzing CCTV image data, and informs the administrator, and the road tunnel disaster prevention facility installation and management guideline (2009.8) is used for such systems. It is recommended that the installed camera be installed at a height of 3m to 3.5m at intervals of 100m to 150m. Road maintenance management organizations such as Korea Expressway Corporation and the Korea Land Management Agency have installed and operated some video image detection systems in tunnels, which are expected to expand gradually.

1 is a diagram illustrating a detection range and a distance of a camera according to a road tunnel disaster prevention facility guideline. As can be seen in Figure 1, the camera is installed at a predetermined interval in the tunnel, it can be seen in the situation room to check the state of Yugo within the detection range of each camera through CCTV images.

Research on object tracking can be divided into area based tracking, feature point based tracking, model based tracking and filter based tracking. However, all of these methods tend to be difficult to respond quickly to the notice situation because they cannot grasp the type of the object, and there is a problem that the object tracking fails when the overlapping phenomenon occurs in detecting the notice situation.

For the same reason as above, there is a need for an algorithm that can classify objects and detect and track object overlap.

The present invention is proposed to improve such a conventional problem,

The purpose of this study is to provide a method to accurately classify the types of objects in detecting the presence of tunnels and to detect and track the overlap of objects.

In order to achieve the above object, in the object recognition method of the tunnel note system of the present invention, a plurality of objects photographed through a camera are classified into photographing distance sections and labeled as rectangles, and the horizontal and vertical ratio information of the labeled objects ( Hereinafter, as a method of recognizing an object in a tunnel notice system for storing all modeling information) and detecting a moving object through the camera image,
(a) labeling a rectangle along an outer edge of each of at least one moving object detected by the camera; And
(b) detecting a ratio of the horizontal and vertical values of each of the one or more moving objects labeled in step (a), and classifying the type of the moving object by comparing with the modeling information;
After classifying at least one kind of the moving object, a horizontal and vertical labeling ratio (first ratio) of the moving object included in the current frame image including the moving object and the moving object included in the previous frame image; When the horizontal and vertical labeling ratios (second ratios) are compared with each other, when at least one of the first ratio and the second ratios does not coincide with the modeling information, overlapping occurs in at least one or more of the moving objects. It is characterized by judging that.

In addition, the horizontal and vertical numerical values of the labeling of the object recognition method of the tunnel note system of the present invention is located at the distance and the location of the separated distance so as to detect the moving object for each section in the range detected in the image. Preferably, the distance is stored in advance according to the type of the moving object, and the minimum value of the distance is preferably based on the coordinates when the entire appearance of the mobile object, which is partially displayed in the image, is displayed for the first time.

The moving object type classification step of the object recognition method of the tunnel note system according to the present invention may include detecting the moving object as a vehicle when the horizontal and vertical ratios of the labeling are 0.4 or more and less than 2.0, and the ratio is 2.0 or more and 6.0. If less, the mobile object is preferably recognized as a person.

delete

According to the object recognition method of the tunnel note system of the present invention,

It is possible to distinguish the types of moving objects by labeling the rectangles along the outer edge of the moving object detected from the image, and calculating the horizontal and vertical ratios of the labeling to determine the range of the moving objects.

In addition, by comparing the labeling ratio of the current frame image with the labeling ratio of the previous frame image, if the two ratios are different from each other, the overlapping of the moving objects can be detected by determining that the overlapping of the moving objects has occurred, and thus continuous tracking can be performed. have.

1 is a view showing the detection range and the distance according to the road tunnel disaster prevention facility guidelines.
2 is a block diagram according to an embodiment of the present invention.
3 is a block diagram of an object classification step according to an embodiment of the present invention.
4 is a conceptual diagram illustrating overlap detection according to an embodiment of the present invention.
5 is an experimental result according to an embodiment of the present invention.

Hereinafter, an embodiment of an object recognition method of a tunnel note system according to the present invention will be described in detail with reference to the accompanying drawings.

2 and 3 are configuration diagrams according to an embodiment of the object recognition method of the tunnel note system of the present invention.

As shown in FIG. 2, in order to apply the object recognition method of the tunnel note system of the present invention, first, a moving object must be detected through a camera image. For this, an image input step 100 and an object candidate detection step 200 are performed. And object detection step 300 will be preceded. Thereafter, the inventors distinguish the types of objects through the labeling step 400 and the object classification step 500, and detect overlapping moving objects in the overlap detection step 600 to enable continuous object tracking. Therefore, the above three steps of the image input step 100, the object candidate detection step 200 and the object detection step 300 are not included in the technical idea of the present invention, but the present invention is a process to be premised in order to apply the present invention. To help understand the invention, the names and definitions of the parts will be briefly described first.

First, the image input step 100 is a step of receiving an image to be processed through the camera of the CCTV. Images can be divided into 320 x 240, 640 x 480, 720 x 480, 1080 x 720, 1280 x 720, etc. according to the resolution of the camera, and the present invention will be described based on the 1280 x 720p image.

Next, the object candidate detection step 200 is a step of performing a preliminary operation to accurately detect the moving object. On the other hand, the type of object described in the present invention is preferably a vehicle, a person to be tracked in the presence situation in the tunnel. In addition, the moving object described in the present invention should not be interpreted in the sense that it is limited to the currently moving object only, it is preferable to include not only the moving object but also the stationary but movable object. In the description of the present invention, the object and the moving object will be used interchangeably because the object is basically movable.

The object candidate detection step 200 may be divided into a background estimation process and a noise removal process. For background estimation, Gaussian mixture model (MoG), Kernel Density Estimate (KDE), and Running Gaussian Average (RGA) algorithms can be used to separate background and objects. have.

In addition, it is possible to apply the Median Filter (TMF), which is a filter that selects a medium value after arranging the values of a given mask area in order of magnitude to remove noise, and expands the morphology of the morphology. (Dilation) can be used to enlarge the object's characteristics and reduce the background, or, conversely, to reduce the object and emphasize the background through erosion.

As such, object candidates are detected through background estimation and noise removal.

Next, in the object detection step 300, the object and the non-object are distinguished through an Adaboost algorithm for learning the object to be detected, and the object is detected from the previously detected object candidates.

The moving object detected through the above process is labeled with a rectangular frame along the outer edge of the object in the labeling step 400, and in the object classification step 500, classifies the object type according to the size ratio of the moving object, If two objects overlap, the overlap detection step (600) detects it so that it can be continuously tracked.

Here, the object classification step 500 may be composed of the section detection 510, the vehicle determination 520, the person determination 530 again as shown in FIG. However, this is not an essential component and the configuration can be added, deleted or changed depending on the object to be detected.

The section detection 510 detects the moving object by a predetermined distance so as to detect the moving object for each section in the range detected in the image. In one embodiment of the present invention 1 section (511): 20m before, 2 sections (512): 20 ~ 50m, 3 sections (513): 50 ~ 100m, 4 sections (514): 100 ~ 150m, 5 sections (515) ): It can be classified after 150m, but the distance of one section or the number of division sections can be adjusted according to the efficiency of sensing and work.

Thereafter, the vehicle recognition 521 or the person recognition 531 is performed through the vehicle determination 520 and the person determination 530 for each section.

Meanwhile, in order to classify and detect overlapping objects, size information data including height values and width values of objects according to a certain distance and object type are preferably stored in advance, and embodiments thereof are shown in Table 1 below. .

Size according to distance and vehicle type [Unit: Pixel]

As an embodiment of the present invention [Table 1] is a vehicle, the object to detect people, the vehicle is again divided by the type of CAR, SUV, TRUCK by the size, each object is shown in the image farther away from the camera Since the size is also reduced, it is preferable to include the horizontal (W: Wide) and vertical (H: Height) data that vary with each distance (10 m interval in [Table 1]). (Note that the horizontal and vertical values are in pixels on a 1280 x 780 p screen.) In addition, the minimum value of the distance (0m) is the first time the entire appearance of an object that is only partially visible in the image is displayed. It is preferable to refer to the coordinates. That is, in the present invention, a plurality of objects (objects such as a vehicle and a person) photographed through a camera are divided into photographing distance sections and labeled as rectangles, and the horizontal and vertical ratio information (hereinafter, modeling information) of the labeled objects is labeled. It is a system that saves all and compares them with moving objects captured by camera images.

Object classification method according to this configuration is as follows.

If an object is detected in the image, the labeling step is performed by labeling a rectangular frame along the outer edge of the moving object in the labeling step 400, and then, in the object classification step 500, the type of the moving object according to the range of the width-to-height value ratio of the labeling. Separate.

At this time, in order to distinguish the types of objects, first, the horizontal and vertical ratios are calculated and stored based on the size information of the stored objects as shown in [Table 1]. When the ratio of the horizontal to vertical values is expressed as an expression, "ratio = vertical / horizontal" becomes, and objects are recognized according to the calculated ratio range.

If the ratio is 0.7 or more and less than 1.2, the moving object is recognized as a falling object. If the ratio is 1.2 or more and less than 2, the moving object is recognized as a vehicle. When the ratio is 2 or more and less than 5, the moving object is a person. To recognize.

The process of detecting overlaps of objects by comparing the sizes and determining the types of objects by judging the proportions are repeated for each section over the above-described sections 1 to 5 and 415. .

For example, if a moving object is detected in 2 sections 512: 20-50m, the object is labeled, and if the width of the labeled object is 105 pixels wide and 98 pixels long, it can be seen that the object is located approximately 30m among the 2 sections. . In addition, if the aspect ratio is calculated according to the above-mentioned formula, the ratio is 98/105 = 0.93, and since the ratio belongs to 0.7 or more and less than 1.2, it can be seen that the moving object is a vehicle and SUV among the vehicles.

On the other hand, the 0m and 150m points in the detection area are the areas where the tracking of the object starts or ends, and if it is out of these two areas, the object leaves the detection area. You can find out.

After classifying the types of moving objects, the overlap detection step 600 may be additionally performed. In the overlap detection step 600, the labeling ratio of the current frame image is compared with the labeling ratio of the previous frame image.

4 is a conceptual diagram illustrating overlap detection according to an embodiment of the present invention. As shown in FIG. 4, in the overlap detection step 600, individual detection of moving objects is performed, and then object overlap occurs after the object overlap occurs, and object overlap occurs after the movement of the object. When the object overlap is separated, the object is overlapped to detect the overlap and robust tracking is possible.

For example, if the labeling ratio of the previous frame was 0.93 and the labeling ratio of the current frame is 0.5, the labeling ratio of each frame is different, and when referring to [Table 1], only the labeling ratio of the previous frame matches [Table 1] (the horizontal width). 105, vertical 98), the ratio of the previous frame can be interpreted as the ratio of the overlapping due to the movement of another object when the moving object is detected in the detection area. On the contrary, if the labeling ratio of the previous frame was 0.5 and the labeling ratio of the current frame is 0.93, the labeling ratio of each frame is different and only the labeling ratio of the current frame matches [Table 1] when referring to [Table 1]. Since there is a ratio of multiple objects overlapping from the beginning within the detection area, it can be interpreted as the case where the overlapping is released due to the movement of each object.
That is, in the present invention, after the step of classifying at least one kind of the moving object, the horizontal and vertical labeling ratio of the moving object included in the current frame image including the moving object (referred to as the first ratio) and before When the horizontal and vertical labeling ratios (called a second ratio) of the moving object included in the frame image are compared with each other, at least one of the first ratio and the second ratio does not match the modeling information. It is determined that an overlapping phenomenon occurs in at least one of the moving objects.

5 is an experimental result according to an embodiment of the object overlap detection and tracking method of the image detection system of the present invention.

For reference, the experiment used Visual Studio 2010 and OpenCV, an open source library for image processing, on an Intel Core2 Quad Q9300 2.5GHz CPU, and based on the 1280 X 720p image, the object overlap detection and tracking method of the image detection system of the present invention. Experiments were performed by applying the algorithm according to the method. The test result may have some errors depending on CPU, SW integrated development environment, library, and image standard, but it will not make a significant difference.

As can be seen in Figure 5, the image output through the camera installed in the tunnel, that is, the original image 700, it can be observed that three vehicles are passing on the right side of the screen, the resultant image to which the present invention is applied In 710, it can be detected that all of the vehicles are detected, labeled, and tracked.

As described above, the present invention can label the moving objects detected in the image and determine the types of moving objects by determining the horizontal and vertical ratio ranges of the labeling, so that the moving objects can be quickly responded even when the tunnel is in a state of absence. In addition, by comparing the labeling ratio of the current frame image with the labeling ratio of the previous frame image, the overlapping of the moving object can be detected, thereby enabling robust tracking.

As mentioned above, although an embodiment of the object recognition method of the tunnel notice system of the present invention has been described, a person having ordinary knowledge in the relevant technical field is within the scope not departing from the spirit of the present invention described in the claims. The present invention may be variously modified and changed by adding, changing, deleting or adding the same, which will also be included within the scope of the present invention.

100: image input step 200: object candidate detection step
300: object detection step 400: labeling step
500: object classification step
510: Detection by section 511: 1 section
512: 2 sections 513: 3 sections
514: 4 sections 515: 5 sections
520: vehicle determination 521: vehicle recognition
530: people judgment 531: people recognition
600: overlap detection step
700: Original video 710: Result video

Claims (5)

A plurality of objects photographed through the camera are subdivided into photographing distance sections to be labeled with a rectangle, and all aspect ratio information (hereinafter, modeling information) of the labeled objects is stored, and then moving objects are detected through the camera image. As an object recognition method of tunnel notice system,
(a) labeling a rectangle along an outer edge of each of at least one moving object detected by the camera; And
(b) detecting a ratio of the horizontal and vertical values of each of the one or more moving objects labeled in step (a), and classifying the type of the moving object by comparing with the modeling information;
After classifying at least one kind of the moving object, a horizontal and vertical labeling ratio (first ratio) of the moving object included in the current frame image including the moving object and the moving object included in the previous frame image; When the horizontal and vertical labeling ratios (second ratios) are compared with each other, when at least one of the first ratio and the second ratios does not coincide with the modeling information, overlapping occurs in at least one or more of the moving objects. Object recognition method of the tunnel yugo system, characterized in that it is determined that.
delete delete delete delete

Images