KR101407394B1 - System for abandoned and stolen object detection - Google Patents

Abstract

The present invention relates to an image processing server for acquiring an image captured by a surveillance camera and a surveillance camera, and judging whether the surveillance camera and the surveillance camera have left the object and whether or not the object is stolen; A monitoring terminal for displaying an image photographed by a surveillance camera, receiving an alarm signal indicating that an object is left and stolen from the image processing server when the photographed object is a neglected object and a stolen object, By analyzing the time-space changes between people and objects in a video taken by a surveillance camera including a communication network relaying data communication between terminals, it is possible to determine whether the object is left unattended or stolen, And it is possible to prevent the object from being stolen.

Description

SYSTEM FOR ABANDONED AND STOLEN OBJECT DETECTION [0002]

More particularly, the present invention relates to a time-first detection method for analyzing a spatial change after performing time change analysis in a complicated video image captured by a surveillance camera such as a CCTV Relates to a system for detecting unauthorized or stolen objects by analyzing a time-space change between a moving person and a suspicious object.

In general, cameras such as CCTV cameras are installed and used for various purposes in public places. CCTV cameras are mainly installed for monitoring purposes, and they are used for shooting and recording various criminal facts including theft crime and sex crime .

In other words, a CCTV camera that photographs a specific area is only for the purpose of photographing the area and storing the captured image. Actually, the surveillance is performed by a monitor monitoring a video shot by the CCTV camera, There is a problem in that, even if a crime occurs, if the monitoring observer neglects the monitoring, the criminal fact can not be recognized immediately.

That is, the conventional CCTV camera system has a problem that it can not directly notify the occurrence of a crime when a crime occurs, and it can only be used as evidence necessary for proving a crime by a victim or an investigation agency after the occurrence of a crime There is a problem.

Patent Document 1; Korean Patent No. 10-0817753 (Mar. 24, 2008)

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problems, and it is an object of the present invention to provide a time-first detection method for analyzing a spatial change in a complex video image captured by a surveillance camera such as a CCTV, And analyzing a time-space change between a suspicious object and a passive object and detecting a stolen object.

According to an aspect of the present invention, there is provided a system for detecting a stolen object and a stolen object, the system comprising: an image processing server for acquiring an image taken by a surveillance camera and a surveillance camera and determining whether the object is stolen or stolen; A monitoring terminal for displaying an image photographed by a surveillance camera, receiving an alarm signal indicating that an object is left and stolen from the image processing server when the photographed object is a neglected object and a stolen object, And a communication network for relaying data communication in the terminal.

According to the present invention, the surveillance camera analyzes the time and space change between a person and an object in an image taken by a surveillance camera, judges whether the object is left unattended or stolen, And it is possible to prevent the object from being stolen.

Brief Description of the Drawings Fig. 1 is a block diagram of a system for detecting stolen goods,
2 is a detailed configuration diagram of an image processing server according to the present invention,
FIG. 3 is an outline view of an unattended water detection by an approach using spatial information,
FIG. 4 is an outline of the detection of an object based on temporal information forming a trajectory, and
FIG. 5 is a view showing a second-stage leftover detection using region characteristics. FIG.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Prior to this, terms and words used in the present specification and claims should not be construed as limited to ordinary or dictionary terms, and the inventor should appropriately interpret the concept of the term appropriately in order to describe its own invention in the best way. The present invention should be construed in accordance with the meaning and concept consistent with the technical idea of the present invention.

Therefore, the embodiments described in this specification and the configurations shown in the drawings are merely the most preferred embodiments of the present invention and do not represent all the technical ideas of the present invention. Therefore, It is to be understood that equivalents and modifications are possible.

First, in order to detect a leftover object, a system for detecting a leftover object and a stolen object according to the present invention comprises three steps: a moving scanning, a temporal change analysis, and a spatial change analysis step. In order to configure an algorithm for detecting a left object, We analyze space and time characteristics of frames in sequence.

In the first stage, the left-object and theft detection system according to the present invention removes the blob detection, the blob classification using the background modeling, and the model for generating the candidate group of the leftover object, and in the second step, Confirmation Check the validity of the background around the developed normal region. Finally, confirm that at least the minimum time object remains before the warning is issued for the neglected object in the third step.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below with reference to the drawings.

As shown in FIG. 1, the unattended object and stolen object detection system according to the present invention includes a surveillance camera 100, an image processing server 200, a monitoring terminal 300, and a communication network 400.

The surveillance camera 100 is installed in a public place such as an airport, a terminal, a shopping center, and photographs a corresponding area. The surveillance camera 100 monitors the time and space between a person and an object The monitoring terminal 300 displays an image taken by the surveillance camera 100, and at the same time, displays the image taken by the surveillance camera 100 through the image processing server 200 When water or theft is detected, an alarm is sounded so that the watchdog can recognize it.

2, the image processing server 200 includes a transmission / reception unit 210, a background removal unit 230, a blob detection unit 240, a blob classification unit 250, a time characteristic analysis unit 260, a spatial characteristic analyzing unit 270, and a region characteristic analyzing unit 280.

2 is a detailed configuration diagram of the image processing server.

The transmitting / receiving unit 210 receives the image captured by the surveillance camera 100 and transmits a result of the image processing performed by the image processing server 200 to the monitoring terminal 300 .

  The background removal unit 230 removes the background using a general central filtering background removal method of changing the background to use the median value of the previous frame in accordance with the interframe space for the total image photographed by the surveillance camera 100 .

In particular, for robustness, real-time processing, and high accuracy, the model incorporates a mixture of filters performed using Gaussian background removal and functions provided in the OpenCV library.

For adaptive region-based background learning and update performance, the model was developed by two different methods based on pixel-based and non-motion based ghost removal and background updates on tracking objects that are continuously deviating.

In addition, the present invention applies heuristic filters using object size and average height to remove any excessively small and large objects, applying the region of interest to each camera to exclude too far regions.

The main control unit 220 detects all the remaining objects after background removal by the background removal unit 230.

On the other hand, the blob detection unit 240 detects an object detected by the main control unit 200 as a blob.

When a schematic diagram of a left water detected by access using the spatial information with reference to FIG More specifically, the main controller 220 detects the respective object (O _i) the blob detection section 240 is the width with respect to the ( (B _i ) including coordinates (C _i ) having a height (h _i ) and a height (h _i ).

By way of reference, all vectors are obtained from a tracker and are configured in a matrix V _A for all blobs.

Each blob is classified into two different vectors by the blob sorting unit 250, which are classified into a stationary bubble V _S and a moving bubble V _M , respectively.

That is, if the blob does not move in the j-th frame, the blob sorting unit 250 stores the blob in the stop blob V _S , and if the blob moves, stores the blob in the move blob V _M .

The maximum number of blobs is limited within the frame.

In every frame, the matrix consists of one new vector.

If the moving distance d _i of the blob B _i between the frame j and the frame j-1 is greater than the frame threshold df _th , the tracker recognizes that the blob passes through the successive frames.

Meanwhile, the spatial characteristic analyzer 260 determines that the center coordinate of the blob (B _i ) is C _i = (x ^j _i , the y ^j _i) one when the moving distance (d _i) is greater than the calculated by the equation below, and that the movement distances (d _i) movement threshold (d _th ^M) blob to a moving object like a pedestrian or car And recognizes the blob as a stationary object if the movement distance d _i is smaller than the stop threshold value d _th ^S.

Even if the blob is classified as a stop blob, the stop blob is not an object if there is a moving object within the radius r _th .

Also, the spatial property analyzer 260 calculates a distance d ^j _{i , i + 1} between two blobs i and i + 1 in a j frame using the following equation.

If the distance d ^j _{i , i + 1} between the two blobs is not less than the radius r _th , then the blob is not classified as a left blob.

For example, if a person disappears from the field of view, the radius r _th remains unchanged.

Another important problem is the possible similarity between the appearance of a person and the background that can lead to an incorrect segment.

Because of occlusion and error detection, each blob loses or misses the deletion.

In this case, the distance ( ^dj _{i , i + 1} ) is calculated by the distance between the movable object and the movable bobbles around the object.

Above all, the distance (d ^j _{i , i + 1} ) may be less than the radius (r _th ) for each frame depending on the density of the crowd.

The complexity of the problem arises from the overlap, the change of light, the distortion of considerable distinction, and the similarity of the emergence of others.

Therefore, Space First Detection (SFD) results in very high false positives.

To solve these problems, the temporal characteristics of a frame in a video sequence must be considered to classify as a latent object such as a stationary object for a relatively long period of time.

In order to determine the time during which the leftover remains without movement, the present invention utilizes the calculation of the number of frames involved in the stationary blob (V _S ).

As shown in FIG. 4, unattended water detection is made based on time information forming a trajectory.

In the figure, the time-to-live (TTL) timer T _i is the number of frames for the blob B _i contained in the stop blob V _S , And so on.

Where f _j is the current frame number and f _j is the first frame number for the blob (B _i ) contained in the stop blob (V _s ).

For any frame, TTL timer of blobs (B _i) (T _i) is updated during the detection time (△ t _i).

As shown in Fig. 4, the blob B _i has a load and moves from a to b, leaving unloaded at time t _j .

At this time, the time characteristic analyzer 270 determines the load as the left-hand side candidate group (B _{i +1} ) for the time critical point (T _th ).

Also, when the TTL timer T _{i +1} is greater than the time threshold T _th , the time characteristic analyzer 270 sets the left candidate residue group (B _{i +1} ) as a negated object in the frame t _j + T _th Classified and accurate alarms occur.

The time critical point T _th is calculated by the following equation.

Where r is the frame rate and τ is the minimum detection time.

(B _{i + 1} ) from the time t _{j + k} to t _{j + k + 1} are overlapped by the bubbles (B _{i +2} ) of the left-hand candidate. This is an example of a false alarm.

If the TTL timer (T _i ) has a small value, it will detect many fixed objects standing on the roadside or sitting on the bench.

Therefore, many false alarms occur when the number of objects increases.

This will require more memory, more computing means, and more security.

If the TTL timer (T _i ) has a large value, the object can not be detected immediately.

More than ever, there is a need for management of latent neglect, which leads to an increase in memory allocation and a significant computation time close to exponentiation.

 On the other hand, an object tracking method for locating the owner of an object is to trace back the neglected object against time.

The owner of a neglected object is usually defined as a person who has left neglected objects in the video image.

 If the host was previously involved in neglect, the search for the owner begins.

If the time characteristic analyzer 270 finds that the master has disappeared from the detection area for a time longer than the TTL timer ( _Ti ), it is classified as unattended, an alarm is sounded, and the master returns to the accommodation , The TTL timer (T _i ) is reset to zero and the alarm is stopped.

In addition, if the candidate group object moves together with the blob B _{i that} is not related to the object, the temporal characteristic analyzer 270 classifies the candidate object as the stolen object.

As shown in FIG. 4, the TTL timer T _{i + 1} of the banned object B _{i + 1} is reset at time t _{j + k + l} because the owner B _i has returned.

Time 't _{j + k + l + m + n'} in the blob (B _{i + 1)} are blobs (B _i) is' allowed to stand water candidate group in the still blobs (V _S) while moving to b` blobs (B _{i + 1} ) Are classified as stolen because they have disappeared.

It is unclear to determine the appropriate time and distance between an object and an object master to classify the object as left untouched.

In addition, the secretion may not have enough space to allow the proper time and distance between the person and the lost burden in the place.

Another problem is that video monitoring can be subjective.

Because each person has different thoughts about neglect, the guards can be suspicious of suspicious images that others have not suspected.

Although SFD and TFD have significant failures to detect neglect or theft, they need to be addressed in a positional relationship.

In step I, neglected goods and stolen goods are classified by SFD and TFD as potential deposits, and in step II, potential deposits are classified as clear neglected or stolen goods.

FIG. 5 shows a two-step water detection using area characteristics.

As shown, there are two areas A and B, and there are three blobs B _i , B _{i +1} , B _{i +2} .

The region attribute analysis unit 280 B _{i + 1} is allowed to stand and then separated into water, B _i is determined to B _{i + 1} while the location where the radius α outside as the left water and theft of water established by the Ⅱ step .

Similarly, when the region B _i is located in the A region while the B _i moves outside the radius a, the region characteristic analyzing unit 280 analyzes the region B _{i +1} as a non-object according to the domain policy.

As shown in FIG. 5, each area R _k has a TTL timer (T ^rg _k ), and the TTL timer (T ^rg _k ) is set through adjustment of several important parameters for each camera and environment.

The TTL timer (T ^rg _k ) is calculated by the following equation.

Where n is the total number of objects in each region R _k during the training time (frames) f _tr over a time threshold T _th , and A (R _k ) is the area of the area `R <sub>k`</sub> .

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, It is to be understood that various modifications and changes may be made without departing from the scope of the appended claims.

100: surveillance camera 200: image processing server
210: Transmitting / receiving unit 220:
230: background removal unit 240: blob detection unit
250: bubble classifying section 260: time characteristic analyzing section
270: spatial characteristic analyzing unit 280: region characteristic analyzing unit
300: monitoring terminal 400: communication network

Claims (10)

A surveillance camera 100;
An image processing server (200) for acquiring an image taken by the surveillance camera (100), and judging whether the surveillance camera (100) is left unattended or stolen;
A monitoring terminal 300 for displaying an image photographed by the surveillance camera 100 and receiving an alarm signal indicating that an object is left and stolen from the image processing server 200 when the photographed object is a neglected object or a stolen object, ); And
And a communication network (400) for relaying data communication between the monitoring camera (100), the image processing server (200) and the monitoring terminal (300)
The image processing server 200
A background removal unit 230 for removing the background through a central filtered background removal method for changing a background to use a median value of a previous frame in accordance with an interframe space for a total image captured by the surveillance camera 100;
A blob detection unit 240 for detecting a blob that displays each object in the background removed by the background removal unit 230;
A blob sorting unit 250 for sorting all the blobs detected by the blob detecting unit 240 into a stop blob and a moving blob;
A spatial characteristic analyzer (260) for calculating a moving distance or a distance between the blobs between the frames and analyzing the spatial characteristics to detect a leftover object or a stolen object; And
And a time characteristic analyzer (270) for calculating a TTL timer (T _i ), which is the number of blobs included in the stop blob, to detect a leftover object or a stolen object by analyzing a time characteristic. Water detection system.
delete The method according to claim 1,
A step of classifying the object or the stolen object as a latent object by the spatial characteristic analyzer 260 and the temporal characteristic analyzer 207 and the step of classifying the latent object into a definite object or stolen object And a region characteristic analysis unit (280) for detecting the characteristic of the stolen object.
The method according to claim 1,
The spatial characteristic analysis unit 260
The blob (B _i) are the coordinates C = _i when _(i x ^j, y ^j _i) the moving distance (d _i) between the frame

Of the detected object.
5. The method of claim 4,
When the moving distance d _i is
And recognizes the blob as a moving object if the moving speed is greater than the moving threshold value ( _dth ^M ) and recognizes the blob as a stationary object if the moving speed is smaller than the stopping threshold ( _dth ^S ).
6. The method of claim 5,
Wherein the spatial characteristic analyzer (260) analyzes the object as a non-object if the object has a moving object within the radius r _th of the object.
The method according to claim 1,
The spatial characteristic analysis unit 260
(D ^j _{i, i + 1} ) between two blobs (i and i + 1) in any one frame j

Lt; / RTI &gt;
Characterized in that if the distance (d ^j _{i, i + 1} ) between the two blobs is not less than the radius r _th , then the blob is not classified as a left blob.
The method according to claim 1,
The time-characteristic analyzer 270 receives the TTL timer T _i

, Wherein f _j is the current frame number and f _k is the first frame number for the blob (B _i ) contained in the stop blob (V _s ).
9. The method of claim 8,
The time characteristic analysis unit 270
TTL timer time threshold (T _th = r × σ; r is the frame rate, τ is the minimum detection time) when greater than, left standing water candidate group (B _{i + 1)} the classified as leaving water in frame t _j + T _th So that an alarm is accurately generated.
10. The method of claim 9,
The time characteristic analysis unit 270
And the alarm is stopped when the host of the neglected object returns to the neglected object by resetting the TTL timer ( _Ti ) to zero.

Images