KR20160093253A - Video based abnormal flow detection method and system - Google Patents

Abstract

The present invention provides video based abnormal flow detection method and system to detect an abnormal flow from a video in real time on the basis of a learning result. The method comprises the steps of: receiving video from a video camera capturing a monitoring area to learn a normal motion flow about the monitoring area for a predetermined time; using a learning result to detect an abnormal flow occurring in the monitoring area from the video of the video camera in real time after learning; and generating an alarm to notify occurrence of the abnormal flow when the abnormal flow is detected, and storing the video including the abnormal flow. A step of learning the normal motion flow comprises the steps of: receiving a video frame from the video camera; dividing the video frame into a plurality of sub blocks; acquiring a motion vector from each pixel of the video frame; using the acquired motion vector to update a motion direction histogram of each sub block; and using the motion direction histogram to generate a motion direction map. The motion direction map is generated after the motion direction histogram is updated for a predetermined period.

Description

TECHNICAL FIELD [0001] The present invention relates to an image-based abnormal flow sensing method,

When a crowd or a vehicle showing a constant flow shows abnormal flow different from usual due to the occurrence of an accident / accident when a crowd or a vehicle frequently watches a traffic passage through the CCTV, The present invention relates to a system for analyzing a video image in real time and detecting such an abnormal flow and providing an alarm, and a method thereof.

Conventional CCTV system is a passive monitoring system, which relied entirely on the interpretation of images obtained from CCTV cameras. Recently, as computer processing capability and computer vision technology developed, CCTV camera images are analyzed in real time Intelligent video surveillance systems that detect events and provide alarms to surveillance personnel are being introduced.

Most intelligent video surveillance systems and related technologies that have emerged to date focus on detecting events generated by individual objects through image analysis. That is, the system detects and tracks individual moving objects from the video image, and determines whether the object generates specific events such as area invasion based on object tracking results such as moving trajectories and the like. Examples of techniques for detecting and tracking individual objects in a video image are disclosed in Korean Patent No. 10-0312427 entitled " Method and apparatus for tracking moving objects using template and Kalman filter in video image ", Korean Patent No. 10-0364582 Moving object tracking / monitoring system ", Korean Patent Registration No. 10-0635883 "real-time object tracking device ", and the like.

Since the intelligent video surveillance system and techniques focus on detecting and tracking individual moving objects, it is possible to utilize it in an environment in which a relatively small number of moving objects are dispersed and moved. However, in an environment where a crowd or a vehicle is frequently crowded and moves It is difficult to utilize.

Also, in the above-mentioned intelligent video surveillance system and technology, the contents of the event to be detected must be clearly described in the behavior of the individual object. Therefore, it is not suitable for detecting an event such as "occurrence of an unspecified event / accident ".

Korean Patent Publication No. 10-2001-0057768 (July 2001) Korean Patent Publication No. 10-2001-0000107 (2001.01.05) Korean Patent Publication No. 10-2002-0097449 (December 31, 2002)

The present invention relates to a method and system for monitoring a traffic flow in which a crowd or a vehicle frequently moves through CCTV, learning a normal flow of a crowd or a vehicle through a video image for a predetermined period of time, The present invention provides a system and method for detecting abnormal flow in real time from a video image based on a learning result when a vehicle or the like shows an abnormal flow different from usual.

A method of detecting an abnormal motion based on an image according to an embodiment of the present invention includes: receiving a video image from a video camera that captures a surveillance region and learning a normal motion flow for a surveillance region for a predetermined time; Sensing an abnormal flow occurring in the monitoring area in real time using the learning result from the image of the video camera after the learning; Generating an alarm when the abnormal flow is sensed and notifying that an abnormal flow has occurred and storing a video image in which an abnormal flow has occurred, wherein learning the normal motion flow comprises: receiving a video frame from the video camera step; Dividing the video frame into a plurality of sub-blocks; Obtaining a motion vector for each pixel of the video frame; Updating a motion direction histogram of each sub-block using the obtained motion vector; Generating a motion direction map using the motion direction histogram, wherein the motion direction map is generated after the motion direction histogram is updated for a specified period of time.

The motion direction map is generated by using the motion direction histogram to generate the motion direction angle values of the local peaks of the motion direction histogram corresponding to the specific subblock to the element value of the motion direction map corresponding to the subblock And the like.

The detecting of the abnormal flow may include: receiving a video frame from the video camera; Extracting a foreground region from the input video frame using a background model of the surveillance region being photographed by the video camera; Obtaining a motion vector for each pixel in the foreground region; Classifying each pixel in the foreground region into a motion pixel and a non-motion pixel; Determining whether motion pixels are abnormal using the motion direction map; Updating an anomaly value for all pixels of the video frame; Obtaining an abnormal flow generation region from pixels whose unsteadiness value is equal to or higher than a specified threshold value; And determining whether an abnormal flow has occurred from the size or area of the abnormal flow generation region.

Also, classifying each pixel in the foreground region into motion pixels and non-motion pixels classifies the pixel as a motion pixel if the size of the motion vector of the pixel is greater than or equal to a specified threshold, and if the size of the pixel's motion vector is less than the specified threshold And is classified into motion pixels.

The determination of whether or not the motion pixels are abnormal by using the motion direction map is performed by comparing the direction angle value of the motion vector of each motion pixel with the motion direction angle values registered in the motion direction map corresponding to the motion pixel And when the similarity is low, it is determined as " abnormal ".

Also, updating the anomaly value for all pixels of a video frame may increase the anomaly value if the particular pixel is a " motion pixel and anomalous " or a " non-motion pixel " And if the pixel belongs to the " pixel not belonging to the foreground region ", the non-uniformity value is reduced.

According to an aspect of the present invention, there is provided an image-based abnormal flow detection system comprising: a video camera for photographing a surveillance area; And a control unit that receives a video image from the video camera and learns a normal motion flow for a surveillance area, and detects an abnormal flow occurring in the surveillance area in real time using the learning result from the video camera, A normal motion flow learning unit for learning a normal motion for a surveillance region from a video image input from a video camera for a predetermined time and providing a motion direction map as a learning result; An abnormal flow sensing unit for sensing in real time an abnormal flow occurring in a monitoring area from the image of the video camera using the motion direction map and providing an abnormal flow occurrence and an abnormal flow occurrence area as a detection result; And an alarm generating unit for generating an alarm indicating that abnormal flow is detected by the user through the video or audio output device when an abnormal flow is detected from the abnormal flow sensing unit.

The image processing apparatus stores video image data for a predetermined period of time before and after detecting an abnormal flow when detecting an abnormal flow occurring in the surveillance region through the abnormal flow sensing unit, A video / event storage unit for storing link information on the stored video image data in an event DB; A video / event search unit for searching the event database for events matching the search condition specified by the user and displaying the event on the screen; And a video / event playback unit for taking the video image data linked with the event selected by the user from the storage device and reproducing the extracted video / image data on the screen.

In addition, the event information for the detected abnormal flow includes at least one of a time at which the abnormal flow occurred, an ID of the camera that sensed the abnormal flow, and an area where the abnormal flow occurred in the image.

In addition, when an abnormal flow occurring in the surveillance area is detected through the abnormality detection unit of the image processing apparatus, the external device connected through various wired / wireless communication lines detects whether abnormal flow is detected, event information on the detected abnormal flow, And a communication unit for communicating at least one of video image data with an external device.

As described above, according to the image-based abnormal flow detection system and method according to the present invention, abnormal flows generated by events / accidents in a crowded passage or a frequent moving vehicle are detected in real time through CCTV, To help the surveillance officer quickly identify and respond to incidents / incidents in public places.

1 is a configuration diagram of an image-based abnormal flow detection system according to an embodiment of the present invention.
2 is a main operation flowchart of an image processing apparatus according to an embodiment of the present invention.
3 is a flow chart for normal motion flow learning.
Figure 4 is a view of a typical motion vector direction histogram.
5 is a flowchart illustrating an abnormal flow detection through image analysis.

1 is a configuration diagram of an image-based abnormal flow detection system according to an embodiment of the present invention.

1, the abnormal flow detection system according to an embodiment of the present invention basically includes a video camera 101 and an image processing apparatus 102. Incidentally, the image processing apparatus 102 includes various external apparatuses 103, respectively.

The video camera 101 may be a general fixed camera or a PTZ camera capable of zooming, and may be a camera that outputs an analog video signal such as NTSC / PAL according to a method of outputting a video signal, Or the like, and outputs the digital video signal through a digital interface. When the video camera 101 is a PTZ camera, the video camera 101 must be in a fixed state during image analysis through the image processing apparatus 102.

The image processing apparatus 102 receives a video signal from the video camera 101 and acquires a series of digital video image data. The video processing apparatus 102 analyzes the video image in real time to detect an abnormal flow of a crowd or a vehicle in the monitoring area, And alarm occurrence and video image recording when flow detection is performed. Also, the image processing apparatus 102 communicates with various external devices 103 such as a portable terminal, a DVR, and a central control device through various wired / wireless communication means, if necessary.

The image processing apparatus 102 includes a video acquiring unit 1021, a normal motion flow learning unit 1022, an abnormal flow sensing unit 1023, an alarm generating unit 1024, a screen display unit A video / event playback unit 1028, a user command input unit 1029, and a communication unit 1030 with an external device. The video / event storage unit 1026, the video / event storage unit 1026,

The video acquisition unit 1021 receives a video signal from the video camera 101 and acquires a series of digital video frame data.

The normal motion flow learning unit 1022 learns a normal motion flow generated by moving objects such as a crowd or a vehicle in a normal monitoring area from the acquired video data. Learning of the normal motion flow is performed for a predetermined time before abnormal flow sensing is performed in the abnormal flow sensing unit 1023. [

Based on the information of the normal motion flow learned by the normal motion flow learning unit 1022, the abnormal flow detection unit 1023 analyzes the video image of the monitored region in real time to detect abnormal flow of moving objects such as a crowd or a vehicle Abnormal Flow).

When the abnormal flow sensing unit 1023 senses an abnormal flow, the alarm generating unit 1024 outputs an alarm signal through a monitor or a speaker to inform the surveillant that an abnormal flow has occurred, And sends an alarm signal to the external system 103 connected to the external device 102.

The screen display unit 1025 displays various information such as a real-time / playback video image, an alarm message, and a system operation status through a monitor.

The video / event storage unit 1026 stores video data for a predetermined period of time before and after an abnormal flow occurs when the abnormal flow detection unit 1023 detects an abnormal flow. Also, information on the detected event is registered in the event DB and linked with the stored video data.

The video / event search unit 1027 searches the event DB for events that satisfy the search condition specified by the user and displays the search result on the screen.

The video / event playback unit 1028 fetches the video data linked with the event selected by the user among the events searched by the video / event search unit 1027 from the storage device and displays the video data on the screen.

The user command input unit 1029 receives various setting / control commands from the user and transfers the commands to the related module.

The communication unit 1030 communicates with the external device via the various communication means, such as an alarm message, video / event data, etc., to the external system, or receives the setting / control command from the external system.

2 is a main operation flowchart of the image processing apparatus 102 according to an embodiment of the present invention.

In step S101, the image processing apparatus 102 acquires a video image from the video camera 101. [

In step S102, the image processing apparatus 102 learns a normal motion flow generated by moving objects such as a crowd or a vehicle in the normal monitoring area from the video image. Learning of the normal motion flow is performed for a sufficiently long time before performing abnormal flow sensing in step S103, and acquires a motion direction map as a learning result.

In step S103, the abnormal flow sensing unit 1023 of the image processing apparatus 102 extracts an abnormal flow of moving objects such as a crowd or a vehicle from the video image using the motion direction map acquired in step S102, In real time.

In step S104, the image processing apparatus 102 generates an alarm through a monitor screen or a speaker when an abnormal flow is detected, thereby notifying the surveillant that an abnormal flow has occurred.

In step S105, the communication unit 1030 with the external device of the image processing apparatus 102 transmits alarm and detected abnormal flow event information to various external devices 103 through various communication means, if necessary. Various information can be transmitted according to the type of the event, the location and time of the event, and the related video data from the signal indicating whether or not the event is detected.

In step S106, the video / event storage unit 1026 of the image processing apparatus 102 stores relevant video images and event information in the local storage when an abnormal flow event is detected. The video image data is stored for a designated time before and after the event detection point, and is stored using a compression method such as MPEG4 / H.264. The related event information is stored in the event database (DB) so that it can be retrieved afterwards.

In step S107, the video / event searching unit and the reproducing units 1027 and 1028 of the image processing apparatus 102 search for the stored events according to the request of the user and reproduce the related video images. The image processing apparatus 102 has a list of recently detected events so that when the user selects a specific event item in the list, the related video image is instantly reproduced. When the user inputs a search condition (event type, occurrence time, etc.) for an event to be searched in the image processing apparatus 102 and issues a search instruction, the image processing apparatus 102 searches for events The list is displayed on the screen, and when the user selects a specific event item in the list, the related video image is reproduced.

3 is a flowchart illustrating a normal motion flow learning operation corresponding to step S102 of FIG.

In step S201, the video processing apparatus 102 receives a video frame from the video camera 101. [

In step S202, the image processing apparatus 102 equally divides the input video frame into sub-blocks of the same size. Usually, the size of the sub-block is set to about 4 x 4.

, 이전 프레임에서의 픽셀 좌표를

라고 하면, 현재 프레임에서 픽셀 p의 모션 벡터 v(p)는 하기 수학식 1과 같이 주어진다.In step S203, the image processing apparatus 102 obtains a motion vector for each pixel of the input video frame. When a point P in the space is moving, the pixel coordinates in the corresponding current frame P

, The pixel coordinates in the previous frame

, The motion vector v (p) of the pixel p in the current frame is given by the following equation (1).

Given the current frame and its previous frames, there are various ways to obtain the motion vector of each pixel in the current frame. For example, J. L. Barron, D. J. Fleet, and S. Beauchemin in "Performance of Optical Flow Techniques" (International Journal of Computer Vision, Vol. 12, No. 1, 1994) disclose various methods for obtaining motion vectors.

In step S204, the image processing apparatus 102 updates the motion vector direction histogram for each sub-block. A motion vector orientation histogram is a one-dimensional array composed of N bins. The value of each bin represents the frequency of appearance of a motion vector having a direction angle belonging to a specific range. Each sub-block is assigned a motion vector direction histogram, and the motion vector of each pixel in the sub-block is used to update the motion vector direction histogram assigned to that sub-block. Assuming that the motion vector direction histogram of the specific subblock B is H, the i-th bean of the histogram H is H [i], and the motion vector of a specific pixel p belonging to the subblock B is v (p) Is updated as shown in the following equation (2).

는 v(p)의 방향각,

는 v(p)의 크기,

은 임의의 임계값을 나타낸다. 모션 벡터 방향 히스토그램의 각 빈은 학습이 시작되기 전에 0으로 초기화되어 있다고 가정한다. In Equation (2)

Is the direction angle of v (p)

Is the size of v (p)

Represents an arbitrary threshold value. It is assumed that each bean in the motion vector direction histogram is initialized to zero before learning begins.

보다 작은 픽셀의 경우에는 모션이 발생하지 않은 것으로 간주하고 히스토그램 업데이트 시 사용되지 않는다.If the size of v (p)

For smaller pixels, motion is considered not to occur and is not used when updating the histogram.

In step S205, the image processing apparatus 102 checks whether the designated learning time has elapsed. If the designated learning time has not elapsed, steps S201 to S204 are repeatedly performed. The learning time is usually set from one day to several days so that the image processing apparatus 102 can sufficiently learn the normal motion flow within the surveillance region.

When the designated learning time has elapsed, the image processing apparatus 102 generates a motion direction map (Motion Direction Map) using the accumulated motion vector direction histogram in step S206. The motion direction map is a two-dimensional array formed by extracting representative motion direction angle values of each sub-block. The value of each bin in the motion vector direction histogram of a specific sub-block means the number of times a motion vector having a directional angle within a range corresponding to the bin during the learning time has occurred in the sub-block. Therefore, the local peak of the motion vector direction histogram of a specific sub-block indicates a typical direction angle of motion generated during the learning time in the sub-block.

Figure 4 shows the appearance of a typical motion vector direction histogram and corresponding local peaks.

라고 하면, 모션 방향 맵 M의 (i,j)에 위치하는 요소 M(i,j)의 값은 하기 수학식 3과 같이 주어진다.The number of local peaks in the motion vector direction histogram may be one or several. For example, if the monitoring area is a path that allows only unidirectional motion, only one representative motion exists, so that the motion vector direction histogram of most subblocks will have only one local peak. If the surveillance region is a path through which bidirectional motion is allowed, there are two representative motions, so that the motion vector direction histogram of most subblocks will have two local peaks. The motion vector direction histogram of the subblock B (i, j) located at (i, j) in the evenly divided image is referred to as H (i, j) and the bin corresponding to the local peak of H The index value

, The value of the element M (i, j) located at (i, j) in the motion direction map M is given by the following equation (3).

는 임의의 임계값으로, 노이즈에 해당하는 로컬 피크는 고려 대상에서 제외시키기 위해 사용된다. 상기 수학식 3을 통해 생성된 모션 방향 맵은 도 2의 단계(S103)에서 영상 분석을 통한 이상 흐름 감지 시 사용된다.In Equation (3)

Is an arbitrary threshold value, and the local peak corresponding to the noise is used to exclude from consideration. The motion direction map generated through Equation (3) is used in abnormal flow detection through image analysis in step S103 of FIG.

FIG. 5 is a flowchart illustrating an abnormal flow detection process through image analysis corresponding to step S103 of FIG. 2. Referring to FIG.

In step S301, the video processing apparatus 102 receives a video frame from the video camera 101. [

In step S302, the image processing apparatus 102 extracts the foreground region from the input video frame using the background model for the surveillance region image. Background Model refers to a result of modeling the brightness distribution of pixels belonging to a terrain / object (normally, a terrain / object without movement for a long time) constituting the background of the surveillance region in the image. The foreground region is a collection of foreground pixels. Foreground pixels represent pixels having brightness values different from background models in an input video image, and are typically caused by movement of an object, illumination change, or the like. In the case of the background model, it is not necessary to update the background model if the brightness of the pixels of the background area is always constant over time. In most outdoor environments, however, the brightness of the pixels of the background area . Also, the arrangement of objects in the surveillance area may change even if it is indoors. Therefore, it is necessary to update the periodic background model in order to maintain the accurate background model. There are various methods of creating and updating a background model and extracting foreground regions from an input video frame using a background model. For example, C. Stauffer, W. Grimson, "Learning Patterns of Activity Using Real-Time Tracking" (IEEE Trans. On PAMI, Vol. 22, No. 8, 2000), R. Collins, A. Lipton, H. Refer to Fujiyoshi, T. Kanade, "Algorithms for Cooperative Multisensor Surveillance" (Proceedings of the IEEE, Vol. 89, No. 10, 2001), Korean Patent No. 10-0364582 "Multiple Moving Object Tracking / Detection System" do it.

In step S303, the motion vector generation unit 102a of the image processing apparatus 102 acquires a motion vector for each pixel in the extracted foreground region.

In step S304, the image processing apparatus 102 classifies the pixels in the extracted foreground region into a motion pixel and a non-motion pixel. If the magnitude of the motion vector is greater than or equal to a certain threshold value, it is classified as a motion pixel. If the motion vector is less than or equal to the threshold value, the threshold value is set to a very small value. When pixels in the foreground region are classified into motion pixels and non-motion pixels, the motion pixels indicate the foreground pixels detected by the instantaneous motion of the currently moving object, and the non-motion pixels cause other causes Gt; pixels < / RTI > The reason why motion pixels are generated is that when an object that is moving is stopped, an object not belonging to the background is abandoned, an object belonging to the background is removed, And a case in which the pixel brightness becomes different from each other.

In step S305, the image processing apparatus 102 updates the background model. The updating of the background model is performed periodically, but not necessarily every video frame. Pixels classified as motion pixels are excluded from updating when the background model is updated. Pixels classified as motion pixels do not provide information for updating the background model since they are pixels from moving objects rather than background objects.

로 주어졌을 때, 상기 모션 픽셀 p가 하기 수학식 4를 만족하면 비정상이라고 판별한다.In step S306, the image processing apparatus 102 determines whether the motion pixels are abnormal using the motion direction map. If the direction of the motion vector corresponding to the specific motion pixel is different from a normal motion direction registered in the motion direction map by a predetermined amount or more, the motion pixel is determined to be abnormal. That is, the motion vector of the motion pixel p belonging to the sub-block B (i, j) is v (p) and the motion vector of the element M (i, j) of the motion direction map corresponding to the sub- A set of direction angles

, It is determined that the motion pixel p is abnormal if it satisfies the following equation (4): " (4) "

는 사용자 지정 임계치이다.In Equation (4)

Is a user-specified threshold.

In step S307, the image processing apparatus 102 updates the abnormality value for each pixel in the image. Each pixel in the image has a corresponding Abnormality Value and initially has a value of zero. The unsteadiness value of each pixel is used to determine how much abnormal motion has occurred in that pixel in a short period of time. If the unsteady value of a specific pixel p is E (p), E (p) is updated as shown in the following equation (5).

: p가 넌모션 픽셀인 경우

: If p is a non-motion pixel

: p가 모션 픽셀이고 정상이거나, 전경 영역에 속하지 않은 경우

: If p is a motion pixel and is normal or does not belong to the foreground region

,

,

는 0보다 큰 값을 갖는 상수이며, 보통

가 되도록 설정한다. 만약 E(p)를 업데이트한 결과가 0보다 작은 값이 되면 0으로 유지한다.In Equation (5)

,

,

Is a constant with a value greater than 0,

. If the result of updating E (p) is less than 0, it is kept at 0.

In the method of updating the abnormality value, although the same abnormality value is updated irrespective of whether the pixel p is a motion or non-motion pixel, an abnormality value updating according to the type of the pixel p as shown in the following equation (6) or .

: p가 모션 픽셀이고 정상이거나, 모션 픽셀이 아닌 경우

: If p is a motion pixel and is normal or is not a motion pixel

: p가 넌모션 픽셀이 아닌 경우

: If p is not a non-motion pixel

Equation (6) or (7) can be used to distinguish the cause of the occurrence of the abnormal motion.

In step S308, the image processing apparatus 102 extracts pixels having an abnormality value of a predetermined threshold value or more. The threshold value controls the period during which abnormal motion occurs in a specific pixel.

In step S309, the image processing apparatus 102 determines whether an abnormal flow occurrence area is generated and an abnormal flow has occurred. The abnormal flow generation region is basically an area consisting of the pixels extracted in step S308. The very small area among the spatially independent areas is removed and removed as a noise corresponding area, To obtain the final abnormal flow generation region. If the area occupied by the abnormal flow generation region in the entire image is greater than or equal to a specific threshold value, it is determined that an abnormal flow has occurred.

Claims (10)

Receiving a video image from a video camera capturing a surveillance region and learning a normal motion flow for a surveillance region for a predetermined period of time;
Sensing an abnormal flow occurring in the monitoring area in real time using the learning result from the image of the video camera after the learning;
Detecting an abnormal flow, generating an alarm to indicate that an abnormal flow has occurred, and storing a video image in which an abnormal flow has occurred,
Wherein learning the normal motion flow comprises:
Receiving a video frame from the video camera;
Dividing the video frame into a plurality of sub-blocks;
Obtaining a motion vector for each pixel of the video frame;
Updating a motion direction histogram of each sub-block using the obtained motion vector; And generating a motion direction map using the motion direction histogram,
Wherein the motion direction map is generated after the motion direction histogram is updated for a specified period of time. 3. The method of claim 2, wherein generating the motion direction map using the motion direction histogram comprises:
Wherein the motion direction angle values of local peaks of a motion direction histogram corresponding to a specific sub-block are registered as element values of a motion direction map corresponding to the sub-block. The method of claim 1, wherein sensing the abnormal flow comprises:
Receiving a video frame from the video camera;
Extracting a foreground region from the input video frame using a background model of the surveillance region being photographed by the video camera;
Obtaining a motion vector for each pixel in the foreground region;
Classifying each pixel in the foreground region into a motion pixel and a non-motion pixel;
Determining whether motion pixels are abnormal using the motion direction map;
Updating an anomaly value for all pixels of the video frame;
Obtaining an abnormal flow generation region from pixels whose unsteadiness value is equal to or higher than a specified threshold value; And determining whether an abnormal flow has occurred from the size or area of the abnormal flow generation region. 4. The method of claim 3, wherein classifying each pixel in the foreground region into a motion pixel and a non-
And classifying the motion vector into a non-motion pixel if the magnitude of the motion vector of the pixel is greater than or equal to the threshold value, and classifying the motion vector into the non-motion pixel if the magnitude of the motion vector of the pixel is less than the predetermined threshold value. 4. The method of claim 3, wherein determining whether the motion pixels are abnormal using the motion direction map comprises:
The similarity between the direction angle value of the motion vector of each motion pixel and the motion direction angle values registered in the motion direction map corresponding to the motion pixel is compared, and if the similarity is low, it is judged as "abnormal" Flow sensing method. 4. The method of claim 3, wherein updating the anomaly value for all pixels of the video frame comprises:
If the specific pixel is a "motion pixel and an abnormal" or a "non-motion pixel", the abnormality value is increased and if the specific pixel is a "motion pixel and normal" or a "pixel not belonging to the foreground area" Based abnormal flow sensing method. A video camera for capturing a surveillance region and a video image from the video camera, and learning a normal motion flow for the surveillance region. Then, using the learning result from the video camera, And an image processing device for detecting an alarm and giving an alarm,
The image processing apparatus comprising:
A normal motion flow learning unit that learns normal motion for a surveillance region from a video image input from the video camera for a predetermined time and provides a motion direction map as a learning result;
An abnormal flow sensing unit for sensing in real time an abnormal flow occurring in a monitoring area from the image of the video camera using the motion direction map and providing an abnormal flow occurrence and an abnormal flow occurrence area as a detection result; And
And an alarm generation unit for generating an alarm indicating that abnormal flow is detected to the user through the video or audio output apparatus when an abnormal flow is detected from the abnormal flow detection unit. The image processing apparatus according to claim 7,
When the abnormal flow occurring in the surveillance area is detected through the abnormal flow sensing unit, the video image data for a predetermined time before and after detecting the abnormal flow is stored in the storage device, and event information on the sensed abnormal flow and the stored video image data A video / event storage unit for storing link information with the event database;
A video / event search unit for searching the event database for events matching the search condition specified by the user and displaying the event on the screen; And
Further comprising a video / event playback unit for taking video image data linked with an event selected by the user from the storage device and reproducing the video image data on the screen. The method of claim 7,
Wherein the event information for the sensed abnormal flow includes at least one of a time at which the abnormal flow occurred, an ID of the camera that sensed the abnormal flow, and an area where abnormal flow occurred in the image. The image processing apparatus according to claim 7,
When an abnormal flow occurring in the surveillance area is sensed through the abnormal flow sensing unit, an external device connected through various wired / wireless communication lines detects whether abnormal flow is detected, event information on the detected abnormal flow, Further comprising a communication unit for communicating with at least one external device that transmits at least one image.

Images