KR101454644B1 - Loitering Detection Using a Pedestrian Tracker - Google Patents

Abstract

A loitering detection apparatus using a pedestrian tracker according to an embodiment of the present invention includes: a camera unit (110) for photographing a predetermined area and outputting a captured image; and a main server unit (130) for comparing a loitering value while a pedestrian loiters in a set detection area and a threshold value for the time of loitering, while setting the detection area in the image received from the camera unit (110) and the threshold value for the time of loitering within the detection area, and for generating an alarm.

Description

[0001] The present invention relates to a pedestrian tracker,

The present invention relates to a method for detecting a tramp using a pedestrian tracker.

Generally, a network surveillance system obtains an input image mainly by using a camera. As described in the background art of the prior art publication No. 10-2011-0111106, the types of cameras are classified into a box type camera and a PTZ type camera.

Box type cameras mainly transmit fixed image scenes in fixed position. Therefore, in the image scene sent out from the box type camera, the user can designate the desired area in the fixed scene through the polygon line drawing for object tracking and looping.

Accordingly, there arises a problem that a threshold value for the moving time of the moving object must be specified in order to distinguish between a normal situation and an abnormal situation.

Also, there is a problem in that the user's line drawing process is manually required for tracking and looping of objects related to image scenes not set as presets.

In the absence of such a process, the PTZ type camera performs tracking and looping of the object through the full range search on all objects generated in the Undefined Preset Region, or does not perform its function at all.

In addition, a typical CCTV used for object tracking and reuter detection conventionally expresses a motion of an object moving in an image in a two-dimensional manner, so that a shape such as rotation, extinction, overlapping, and size conversion of a moving object is expressed There is a problem that is not suitable. In addition, when there are many objects in the image, there is a problem that the amount of motion prediction and compensation is excessively increased.

Disclosure of Invention Technical Problem [8] Accordingly, the present invention has been made in order to solve the problems of the prior art as described above, and it is an object of the present invention to prevent a breakage of a public property, a terrorist attack through a Reuter ring detection in a public area and identify a potentially dangerous person, The present invention provides a method for detecting a thrashing by using a pedestrian tracker.

A method for detecting a sagging using a pedestrian tracker according to an embodiment of the present invention includes steps of setting a surveillance region by designating each corner, setting a threshold value for time of reutering within a set surveillance region, If a pedestrian moving within the surveillance zone is detected, comparing the threshold value for the time of the reuter ring with the value of the reuter ring while the pedestrian walks within the set surveillance zone, if the reuter ring value exceeds a threshold value for the time of the reuter ring Generating an alarm, and setting or adjusting an area of the reuter ring.

In addition, it may include an alert detection algorithm that can detect the tingling of an object in the established surveillance zone or in the area of the set or adjusted reactor ring.

In addition, the pseudo impression detection algorithm measures the pedestrian's rejuvenating value using at least one of the detection of the pedestrian bubble, the detection of the reuter ring, and the detection of the reusing pedestrian within the set surveillance area, And outputting an alarm signal when the threshold for time is exceeded.

In addition, the search algorithm can detect the tingling of the object through location analysis, time change analysis, and spatial change analysis, and can include position analysis, time change analysis, and spatial change analysis in chronological order or random .

In addition, the apparatus for detecting a sling using the pedestrian tracker according to an embodiment of the present invention includes a camera unit 110 for photographing a predetermined area and outputting a photographed image, A main server unit for comparing a threshold value for the time of the reuter ring and a time of the reuter ring while the threshold value for the time of the reuter ring is set within the surveillance region and the surveillance region while the pedestrian is standing within the set surveillance region, (130).

The main server unit 130 includes a sensing area 131 for setting a monitoring area by specifying each corner through an image supplied from the camera unit 110, When a moving pedestrian is detected, the pedestrian compares the measured value of the Reuter ring value with the threshold value of the Reuter ring while the pedestrian is standing within the set monitoring area, and outputs a comparison signal. And an alarm terminal 135 for generating an alarm when an alarm signal is supplied from the main control terminal 133 and the main control terminal 133 for setting or adjusting an area of the reuter ring.

In addition, the main control stage 133 may include an algorithm for detecting the presence of objects in the established surveillance area or in the area of the set or adjusted Reuters ring.

In addition, the pseudo impression detection algorithm measures the pedestrian's rejuvenating value using at least one of the detection of the pedestrian bubble, the detection of the reuter ring, and the detection of the reusing pedestrian within the set surveillance area, And outputting an alarm signal when the threshold for time is exceeded.

In addition, the search algorithm can detect the tingling of the object through location analysis, time change analysis, and spatial change analysis, and can include position analysis, time change analysis, and spatial change analysis in chronological order or random .

In addition, the camera unit 110 may include an RGB-D camera that identifies an object that moves without illumination at night, and recognizes the behavior of a pedestrian when it is a pedestrian.

The method of detecting a sling using a pedestrian tracker according to an embodiment of the present invention can generate an alarm while locating and tracking a moving object when a moving object stays within the FOV of the monitoring camera during an automatically calculated time, There is an effect that it can be prevented in advance.

In addition, the method of detecting a slobber using a pedestrian tracker according to an embodiment of the present invention can prevent a public vandalism and a terrorist attack through a Reuter ring detection in a public area, identify a potentially dangerous person, It is possible to effectively protect the people who are exposed to the unprotected people, thereby remarkably enhancing the stability to the people.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram of an apparatus for detecting tethering using a pedestrian tracker in accordance with an embodiment of the present invention. FIG.
FIG. 2 is a flowchart of a method for detecting a slobber using a pedestrian tracker according to an embodiment of the present invention. FIG.
Figure 3 is a diagram illustrating the use of a central bottom point of a blob to obtain coordinates of each blob in a method of detecting a bloom using a pedestrian tracker in accordance with an embodiment of the present invention.
4 is a diagram illustrating a basic operation of a method for detecting a moving object using time information in a method of detecting a moving object using a pedestrian tracker according to an embodiment of the present invention.
FIG. 5 is a block diagram illustrating a method for detecting an idle behavior in a method of detecting a idle state using a pedestrian tracker according to an exemplary embodiment of the present invention; FIG.
FIG. 6 is a flowchart illustrating a behavior detection method in a method for detecting a sling using a pedestrian tracker according to an exemplary embodiment of the present invention; FIG.
Figures 7a and 7b illustrate the distribution of latency in data in a method for detecting tethering using a pedestrian tracker in accordance with an embodiment of the present invention.

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

Referring to FIG. 1, an apparatus for detecting a tramp using a pedestrian tracker according to an embodiment of the present invention is as follows.

First, an apparatus for detecting a tramp using a pedestrian tracker according to an embodiment of the present invention includes a camera unit 110 and a main server unit 130.

The camera unit 110 photographs a predetermined area and outputs the photographed image. At least one camera may be disposed around a predetermined area so that the camera 110 can capture a predetermined area. In this manner, images captured by the camera unit 110 disposed in the vicinity of the predetermined area can be output in real time.

Although not shown, the camera unit 110 includes a transmitter capable of transmitting an image to be captured in real time, and it is also possible to provide an image photographed using a transmitter to a main server unit 130 to be described later, Of smart devices.

In addition, the camera unit 110 may include an RGB-D camera that identifies an object that moves without illumination at night, and recognizes the behavior of a pedestrian when it is a pedestrian. Accordingly, it is possible not only to easily identify moving objects even in the dark night, but also to accurately recognize the behavior of the pedestrian when the moving object is a pedestrian.

The main server unit 130 receives the image from the camera unit 110 and sets a threshold value for the time of the rejuvenation within the surveillance region and the surveillance region of the image, Compares thresholds for time of Reuter ring and generates an alert. The main server unit 130 may include a sensing region end 131, a main control end 133, and an alarm end 135.

The sensing area stage 131 can set a surveillance area by designating each corner through an image supplied from the camera unit 110. [ The sensing area stage 131 can analyze the image provided in real time and set the respective corners in the provided image, and then set the monitoring area.

The main control stage 133 compares the measured values of the Reuter ring value and the Reuter ring time while the pedestrian is moving within the set monitoring area when the pedestrian moving within the set monitoring area is detected from the sensing area stage 131 And outputs a comparison signal. You can set or adjust the area of the Reuter ring. Such a main control stage 133 may include an algorithm for detecting the presence of objects in a set monitoring area or an area of the set or adjusted reactors ring.

Such a wake-up detection algorithm measures the walking value of a pedestrian using at least one of detection of a pedestrian blob in a set surveillance area, detection of a reuter ring, and detection of a reuter ring pedestrian, The alarm signal can be output.

In addition, it can detect the object's tingling through location analysis, time change analysis and spatial change analysis, and can perform position analysis, time change analysis, and spatial change analysis in chronological order or randomly. A detailed description thereof will be described later.

The alarm terminal 135 can generate an alarm when an alarm signal is supplied from the main control terminal 133. When the alarm signal is supplied to the alarm unit 135, the alarm unit 135 can output an alarm that can be recognized from the outside through the speaker.

Referring to FIG. 2, a method for detecting a sagging using a pedestrian tracker according to an embodiment of the present invention is as follows.

First, a surveillance area is set by designating each corner (S110).

And setting a threshold value for the time of the reuter ring within the set monitoring area (S120). And may include a detection algorithm for detecting a tingling of the object in the surveillance region set therein or in the region of the set or adjusted reacting ring.

If a pedestrian moving within the set monitoring area is detected, the step of comparing the threshold value with the time of the reuter ring value while the pedestrian is standing within the set monitoring area (S130). In this case, the pseudo impression detection algorithm can measure the rejuvenating value of the pedestrian using at least one of detection of a pedestrian bubble, detection of a reuter ring, and detection of a reuter ring pedestrian within a set surveillance area. Such a search algorithm may include detecting the tingling of the object through location analysis, time change analysis, spatial change analysis, and performing the location analysis, time change analysis, and spatial change analysis in chronological order or randomly. A detailed description thereof will be described later.

And generating an alarm when the reutering value exceeds a threshold value for the time of the reutering (S150). Where the presence detection algorithm can generate an alarm if the measured reutering value exceeds a threshold value for the time of the reuter ring.

And setting or adjusting an area of the Reuter ring (S150). A detailed description thereof will be described later.

The basic operation for detecting a moving object using spatial information in an apparatus and method for detecting a sagging using a pedestrian tracker according to an embodiment of the present invention as described above is as shown in FIG.

When three objects Oi, Oi + 1, and Oi + 2 move along the green dotted line, the red dotted line, and the orange dotted line shown in FIG. 3, each object may have a center bottom point coordinate of Cibott, Ci + 1bott, Ci + 2bott have. In addition, we can have the radius of each object space such as ri, r + 1, r + 2.

Assuming that each edge of the surveillance region is Pi, Pi + 1, Pi + 2, Pi + 3, monitoring starts when the center bottom point coordinate of each object is inside the surveillance region, .

That is, when a moving object is identified in a predetermined area, the detected object O _i may be denoted as B _i . At this time, the indicated detection object B _i may include coordinates, width, and height. Here you can convert times from image coordinates to real world coordinates.

Here it is assumed that the lower middle of the blob or the bounding box touches the surface of the earth. The point touching the earth surface can be predicted by using camera calibration.

The center bottom point of the blob can be used to obtain the coordinates of each blob shown in Fig. All vectors of the blob can be obtained from the tracker, and can be constructed within the AB matrix of all blobs.

In addition, each blob can be distinguished by two different vectors, SB and PB. At this time, all the pedestrian blobs can be analyzed to detect the possibility of a hanging risk. When the pedestrian moves within the ROB, the position of each frame can be stored in the pedestrian buffer.

Referring to FIG. 4, the basic operation of the object detection method using time information is as follows.

Basically, it can be judged that the waiting time of the pedestrian exceeds the threshold value. That is, when the threshold value is exceeded, the alarm is immediately started.

At this time, if the object moves out of the region of interest (ROI) for a while, it should be traced in the FOV. It should also be tracked if it disappears briefly in the FOV and then comes back into the ROI or back in the FOV.

To deal with this case, escape times can be allowed to calculate the trace time and escape time threshold.

In order to construct such a detection algorithm, we can analyze the spatial and temporal characteristics of a frame in a captured image. In this algorithm, detection of a moving object can be done through three steps: position analysis, time change analysis, and spatial change analysis. Here, the position analysis, the time change analysis, and the spatial change analysis can be performed in time order, simultaneously or randomly.

That is, as shown in FIG. 4, the object may appear to be outside the RIO with occlusion with other objects during Tiout after being in the ROI for Tiin. At this time, when two objects are divided and again in the area during Tiin, a loitering alarm is generated after the initial threshold Tth, and the alarm can be reset after the time Tout_th has elapsed outside of the overlapping area. In this case, the alarm is not generated because the object is again in the region but has left the area before Tth has passed.

Referring to FIG. 5, a block diagram is shown below for a method for detecting the presence of a moving object.

First, it can include blob detection and pedestrian detection for object identification using the BGS model to generate possible Reutering risks. In this case, the BGS model (Background Subtraction Model) is a preprocessing image processing method that separates a background image and a foreground image from an image. It is assumed that a pixel change is a foreground image, a no change image is a background image, .

In addition, it is possible to separate the background and the object using the Gaussian filtering and the adaptive Gaussian mixture modeling to remove the noise from the external environment. Thus, the detected object can remove the noise of the object using the morphology filter.

In addition, the color-based MeanShift or CAMShift algorithm can be used to initially specify the search window of the object to be traced. Thus, by integrating the mixed Gaussian background extraction and the basic filter, robustness, real-time processing, and high accuracy can be easily performed. In other words, it can be combined with mixed Gaussian background extraction and basic filters to achieve robustness, real-time processing, and high accuracy.

To perform background learning and updating at the dynamic region level, pixel-based and non-motion-based backgrounds can be updated. In addition, ROI can be applied to each camera to exclude areas where sky, lake, wall, etc. are not needed or can not be monitored.

By setting the ROI area in this way, the surveillance area setting and the processing speed can be increased. In addition, the ROI can specify the region and easily apply filtering within that region.

And a heuristic filter such as an average human key and an object size can be applied in order to remove a smaller and larger value than a threshold value. That is, since the distance between the camera and the object can not be known, it is difficult to grasp the size of the object, and since the size of the moving object varies greatly, it is necessary to arbitrarily set the average key and size of the moving person or object as the parameter value . At this time, heuristic filters can be used to input the minimum and maximum length / width of object size as threshold values

Thereafter, the space-time information can be evaluated. The information around the fixed area can be examined to identify the object to which it is attached. That is, the space-time information is processed in the Spatio-temporal analysis shown in FIG. 5, and the situation information is evaluated (in order to distinguish the reutilization object by time / space calculation between the environment parameter and the reference value such as the position / validate.

Finally, you can evaluate the object you are wrestling with to ensure that the object stays as fast as possible before the alarm occurs.

Referring to FIG. 6, the flow chart of the behavior detection is as follows.

First, you can select pedestrian detection and blob. In order to do this, it is possible to periodically analyze one frame at a time and process it. Thus, it is possible to identify the area occupied by moving objects in order to analyze and process one frame at a time.

At this time, foreground images can be separated from the background image to obtain the bodily characteristics of each frame. With this information, only pedestrians can be selected by removing other blobs, such as objects and animals, such as noise, cars, shaking trees, and so on.

Given a list of blobs filled in each frame, control can store the bottom center point of the blob in the space filled by the moving objects in the vector.

The waiting time can then be measured in the ROI. That is, five inputs can be received to measure latency within the ROI. Four may be regions and one may be the bottom central location of the blob.

Four straight lines are obtained from four points P ₁ , P ₂ , P ₃ and P ₄ , and four straight lines can be obtained by the following equations (1) and (2) to determine whether or not the blob is within the area. P ^x _i and P ^y _i are the coordinates of ROI P _i .

Thus, in order to determine whether or not the position is within the ROI, two criteria x and y can be obtained by the following equations (3) and (4). At this time, Z is an indicator for determining whether or not the moving bobbin is inside.

In addition, an associated threshold value can be applied. That is, in each frame, B ⁱ _ped can be constructed to be stored in the order of the coordinates of the pedestrian blob. Each row represents a frame and the coordinates of each block can be stored in two columns.

Where the rows of each matrix represent frames and the coordinates of each blob may be stored in two columns. If the blob B ⁱ enters the region, the position is stored in the first row of B ⁱ _pd . If you wait in the area, you can continue to update the information in the matrix.

If the T ⁱ _in of a given blob exceeds T _th , an alarm may occur because the pedestrian stays in this area for a long time. Also, if the blob leaves the area, the index matrix may turn to zero and remove the blob information from the matrix to eliminate the alarm. In some cases, there may be a problem recognizing pedestrians.

For example, if a pedestrian moves from the edge of an area, if it comes out of the area for a few frames, this blob can not be detected for a while. In this case, even in a dangerous situation, the blob's coordinates may disappear from the matrix due to these actions. To solve this problem, a T ⁱ _out threshold value can be used. If there is no coordinate and goes out of the area, it may not disappear from the buffer afterwards.

FIGS. 7A and 7B show the latency distribution in the PETS 2007 data. At this time, the PETS2007 dataset can be composed of 9 scenarios. Each image of each scenario can be obtained from four different types of camera units 110. [

Thus, a third camera can be used to avoid overlap between pedestrians in a congested scene. S00 and S01-08 can be used in datasets for both fore and aft situations.

That is, the T ⁱ _in of S00-S08 can measure the movement time within the ROI. After gathering a set in each frame, _in the distribution of T ⁱ can be obtained by the probability P (T ⁱ _in). P (T ⁱ _in ) can be obtained by the following equation (5).

In Fig. 7B, T ⁱ _in of S00 is drawn with a blue solid line, and T ⁱ _in S01-S08 is a black dotted line. At this time, the distribution range of T ⁱ _{in in} S00 is from 0 to 153. The mean and standard deviation are 62, 57.6. On the other hand, the distribution range in S01-S08 is from 0 to 454. The mean and deviation are 93.2, 114.

The pedestrian tracking is as follows.

If there is more than one person in the monitoring scene, you must be able to distinguish the other pedestrian blobs to work properly. Accordingly, the coordinates of the pedestrian can be stored in the pedestrian matrix buffer in the order in which they appear, and the x coordinate position of the pedestrian between them can be compared. To facilitate this operation, one part of the trace, such as Kalman filter tracking, can be added.

Also, T ⁱ _in is obtained by the following Equation (6) for the detection of slackness. t ⁱ _in is the time spent _in ROI.

Here, the crisis R is obtained by the following equation (7). N (i) is the number of framing frames and ρ is the ratio for risk calculation.

Thus, in order to distinguish an appearance, a threshold value is a very important variable for identifying an object and detecting a rattling. Thus, two considerations for learning thresholds can be considered to deal with various conditions. P _nor (T ⁱ _in ) and P _abn (T ⁱ _in ) _in normal and abnormal conditions are shown in FIG. 7B. The associated threshold value can be obtained by the following equations (8) and (9).

At this time, T _th can be obtained by the following equations (10) and (11).

Further, the escape time threshold value T out - _th can be obtained by the following equations (12) and (13).

Here, if there is no satisfactory number, it can be obtained by the following expression (14).

epsilon can be determined as the size of the normal dataset and calculated in CDF (cumulative distribution function). This CDF can be obtained by the following equation (15) in the following section [-, T ⁱ _in ].

At this time, the following equation (16) can be used to find the intersection with the X axis.

Here, since the limit on the movement time is not considered, when the value of? Is larger than the cumulative rate, the minimum value of T ⁱ _in as T ^x _in can be obtained by the following equation (17).

Thus, finally, T out - _th can be obtained by the following equation (18).

As described above with reference to Equations 1 to 18, if a moving object is identified, the moving time is stored in PB, and the vector can be continuously updated each time the object moves.

If the object is already saved or newly added, a new record can be added. Thus, if an event is detected after storage, the moving time of the object can be compared.

Also, when a moving object candidate object is identified, it can be compared with AB, and it can be discriminated whether the candidate for the slack object is one of the objects of this vector. If the distance between these two objects is less than the predefined association threshold, the movement time can be updated. If not, the object candidate can be added to the PB as a new object.

As a result, wandering behavior can be detected by comparing movement time with wandering decision policy.

It will be understood by those skilled in the art that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. Therefore, it should be understood that the above-described embodiments are to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than the foregoing description, It is intended that all changes and modifications derived from the equivalent concept be included within the scope of the present invention.

Claims (10)

CLAIMS 1. A method for detecting tethering using a pedestrian tracker,
Setting a surveillance region by designating each corner;
Setting a threshold value for time of reuter ring within a set monitoring area;
Comparing the threshold value for the time of the reuter ring with the value of the reuter ring while the pedestrian is moving within the set surveillance zone if the pedestrian moving within the set surveillance zone is detected;
Generating an alarm if the reutering value exceeds a threshold value for the time of the reutering; And setting or adjusting an area of the reuter ring,
The threshold value can be divided into a normal state (P _nor ) and an abnormal state (P _abn )
The normal situation

, And the abnormal situation is obtained by

And detecting a tingling of the object in the set surveillance region or in the area of the set or adjusted reuter ring. ≪ Desc / Clms Page number 20 >
delete The method according to claim 1,
The search algorithm detects the pedestrian blob in the set surveillance area, detects the reuter ring, and detects the pedestrian using the at least one of the following methods. The measured reuter ring value is measured at the time of the reuter ring And outputting an alarm signal when the threshold value of the pedestrian tracker exceeds a predetermined threshold value.
The method according to claim 1,
The tracking algorithm detects the tingling of the object through location analysis, time change analysis, spatial change analysis, and performs a positional analysis, a time change analysis, and a spatial change analysis in chronological order or randomly. A method of detecting a thrash used.
An apparatus for detecting a trampling using a pedestrian tracker,
A camera unit 110 for photographing a predetermined area and outputting a photographed image;
The threshold value for the time of the reuter ring in the surveillance region and the surveillance region is set in the image by receiving the image from the camera unit 110. While the pedestrian is standing within the set surveillance region, And a main server unit 130 for comparing the values and generating notifications, and the threshold value can be divided into a normal state P _nor and an abnormal state P _abn ,
The normal situation

, And the abnormal situation is obtained by

The main server unit 130 includes a sensing area 131 for setting a monitoring area by designating respective corners through an image supplied from the camera unit 110; If a pedestrian moving within the set monitoring area is detected from the sensing area stage 131, a comparison is made between the measured values of the Reuter ring value and the Reuter ring time while the pedestrian is standing within the set monitoring area, A main control stage 133 for setting or adjusting an area of the reuter ring; And an alarm terminal 135 for generating an alarm when an alarm signal is supplied from the main control terminal 133. The main control terminal 133 controls the operation of the object in the set monitoring area, A device for detecting a sagging using a pedestrian tracker including a sagittal detection algorithm capable of detecting sagging.
delete delete The method of claim 5,
The search algorithm detects the pedestrian blob in the set surveillance area, detects the reuter ring, and detects the pedestrian using the at least one of the following methods. The measured reuter ring value is measured at the time of the reuter ring And outputting an alarm signal when the threshold value is exceeded.
The method of claim 8,
The tracking algorithm detects the tingling of the object through location analysis, time change analysis, spatial change analysis, and performs a positional analysis, a time change analysis, and a spatial change analysis in chronological order or randomly. A device for detecting the sagging used.
The method of claim 5,
Wherein the camera unit (110) includes an RGB-D camera that identifies an object moving without illumination at night and recognizes the behavior of the pedestrian when the pedestrian is a pedestrian.

Images