CN105405150A

CN105405150A - Abnormal behavior detection method and abnormal behavior detection device based fused characteristics

Info

Publication number: CN105405150A
Application number: CN201510689357.XA
Authority: CN
Inventors: 许健; 郑慧; 万定锐
Original assignee: Netposa Technologies Ltd
Current assignee: Netposa Technologies Ltd
Priority date: 2015-10-21
Filing date: 2015-10-21
Publication date: 2016-03-16
Anticipated expiration: 2035-10-21
Also published as: CN105405150B

Abstract

The invention provides an abnormal behavior detection method and an abnormal behavior detection device based fused characteristics. The method comprises steps that, according to a detection tracking processing result of a motion target in a to-be-detected video, a behavior type of the motion target is determined; multi-dimensional characteristics of the motion target are extracted, including a pixel point change degree, a pixel point arrangement tightness degree, an integral shape, a frame image similarity degree, motion characteristics, position characteristics and form characteristics; the multi-dimensional characteristics are analyzed and processed according to a characteristic fusion model corresponding to the behavior type, whether the motion target has abnormal behaviors is determined according to the processing result; according to innovative characteristics of the multiple abnormal behaviors, algorithm robustness and stability can be effectively improved, according to the characteristic fusion model acquired through learning and training large-scale abnormal behaviors, the multi-dimensional characteristics are analyzed and processed, problems of algorithm overfitting and insufficient fitting can be effectively avoided, the method is suitable for multiple types of complex application scenes, time cost and manpower cost are greatly saved, and the method has high popularization values.

Description

Based on anomaly detection method and the device of fusion feature

Technical field

The present invention relates to field of video monitoring, in particular to based on the anomaly detection method of fusion feature and device.

Background technology

At present, a lot of public place such as bank, market, station and traffic intersection is provided with video monitoring apparatus, presets the video image in guarded region, and above-mentioned video image is sent to Surveillance center, form supervisory system for Real-time Collection.But, the monitor task of a lot of reality corresponding in above-mentioned supervisory system still needs more manually completing, and existing video monitoring system is the video image of storage of collected usually, and also can not carry out any process to this video image, therefore these are without explaining that the video image of (namely processing) can only be used as post-mordem forensics usually, cannot give full play to real-time and the initiative of monitoring.

Along with in recent years, the pressure of fight against terrorism and violence progressively increases, detect the ever more important and the technology of reporting to the police seems in real time for the abnormal behaviour occurred in specific public place (as subway station, railway station, market etc.), and above-mentioned abnormal behaviour refer generally to run, unexpected rapidly, fight, behavior or the event such as crowd is restless.

In order to give full play to supervisory system for the above-mentioned real-time of detection abnormal behaviour and the advantage of initiative, correlation technique provides a kind of anomaly detection method, comprise: in all targets, select moving target, then a certain class single features (displacement etc. as movement velocity, moving target size and motion) of moving target is extracted, then analyzing and processing is carried out to this single features, finally judge whether the behavior of this moving target belongs to abnormal behaviour category according to analysis processing result; By the real-time analysis of said method energy, tracking and differentiation monitored object (video image namely gathered), and carry out when superior department or government department when anomalous event occurs pointing out and reporting, for government department, associated safety administrative authority timely decision-making carried out to anomalous event and take correct action to provide support.

Inventor finds under study for action, and above-mentioned single features often exists significant limitation in practical application, easily causes failing to report or reporting by mistake of abnormal behaviour.

Summary of the invention

The object of the present invention is to provide the anomaly detection method based on fusion feature and device, can be more accurate and omnibearing abnormal behaviour to be detected, testing result reliability is higher, and effectively improve robustness and the stability of algorithm, be applicable to Various Complex application scenarios, possess very high promotional value.

First aspect, embodiments provides a kind of anomaly detection method based on fusion feature, comprising:

According to the detecting and tracking result of moving target in video to be tested, determine the behavior type of described moving target;

Extract the various dimensions feature in described moving target; It is multiple that described various dimensions feature comprises in the feature of the following stated moving target: the similarity degree of the intensity of variation of pixel, the arrangement tightness degree of pixel, global shape, two field picture, motion feature, position feature and morphological feature;

According to described behavior type characteristic of correspondence Fusion Model, analyzing and processing is carried out to described various dimensions feature, and judge whether described moving target exists abnormal behaviour according to analysis processing result.

In conjunction with first aspect, embodiments provide the first possible embodiment of first aspect, wherein, the various dimensions feature extracted in described moving target comprises:

Calculate the intensity of variation DisturbRate=FC/FA of the pixel of described moving target in predetermined period; Wherein, FA represents the quantity of all pixels in the foreground area that described predetermined period is corresponding; FC represents the quantity not being in the pixel in described foreground area in foreground area before the certain hour cycle; ;

Calculate the arrangement tightness degree CompactRate=CS/CN of the pixel of described moving target in described predetermined period; Wherein, CN represents that in described foreground area, all four neighborhoods are all in the quantity of the foreground pixel point of described foreground area; CS represents that in CN, all four neighborhoods all exist the quantity of the foreground pixel point of luminance difference;

Calculate the circularity Ω=P/ (2*sqrt (π A)) of the global shape of described moving target in predetermined period; Wherein, A represents the area of described foreground area; P represents the girth of described foreground area;

Calculate the similarity degree AreaSimilarRate=AND/OR of two field picture in described moving target in described predetermined period; Wherein, AND represents that in described foreground area, consecutive frame is all in the number of the pixel of foreground area; Described OR represents the number of the pixel not being in described foreground area in described foreground area in consecutive frame.

In conjunction with first aspect, embodiments provide the embodiment that the second of first aspect is possible, wherein, the various dimensions feature extracted in described moving target comprises:

Extract the motion amplitude of the described moving target in predetermined period in described foreground area; Described motion amplitude comprises: the height of length travel, transversal displacement, current foreground area and the width of current foreground area;

And/or,

Calculate the direction of motion OffRate=Offset/Route of described moving target in described predetermined period; Wherein, Offset represents the actual displacement of described moving target, and this displacement comprises transversal displacement and length travel; Route represents the accumulative stroke of all actual displacements in the foreground area in described predetermined period.

In conjunction with first aspect, embodiments provide the third possible embodiment of first aspect, wherein, the various dimensions feature extracted in described moving target comprises:

Extract the change in displacement value in described moving target between every two two field pictures in predetermined period, and when described change in displacement value is greater than predetermined threshold value, the variation frequency of the position feature of described moving target is added 1, until described predetermined period terminates, using the position feature of the final variation frequency as described moving target.

In conjunction with first aspect, embodiments provide the 4th kind of possible embodiment of first aspect, wherein, the various dimensions feature extracted in described moving target comprises:

Calculate the ratio of width to height of the video area at described video place to be tested;

According to the ratio of width to height of described video area, calculate the ratio of width to height of actual foreground area;

The ratio of width to height of the foreground area of described reality and default the ratio of width to height threshold value are contrasted, and judges the actual form that moving target in the foreground area of described reality is corresponding according to comparing result; Described actual form comprises: single form and many people form.

In conjunction with first aspect, embodiments provide the 5th kind of possible embodiment of first aspect, wherein, according to described behavior type characteristic of correspondence Fusion Model, analyzing and processing is carried out to described various dimensions feature, and judges whether described moving target exists abnormal behaviour and comprise according to analysis processing result:

Extract all proper vectors in described various dimensions feature, all proper vectors are expressed as: F=[F1, F2, F3, F4, F5, F6, F7]; Wherein, F represents the set of all proper vectors; Proper vector F1, F2, F3, F4, F5, F6 and F7 are respectively described intensity of variation, described arrangement tightness degree, described global shape, the similarity degree of described two field picture, described motion feature, above-mentioned position feature and described morphological feature;

Obtain the weight coefficient that in described Fusion Features model, all proper vectors are corresponding, above-mentioned weight coefficient is expressed as: M=[w1, w2, w3, w4, w5, w6, w7]; Wherein, M represents the set of all weight coefficients; Weight coefficient w1, w2, w3, w4, w5, w6 and w7 are respectively weight coefficient corresponding to described intensity of variation, described arrangement tightness degree, described global shape, the similarity degree of described two field picture, described motion feature, above-mentioned position feature and described morphological feature;

According to formula described all proper vectors and corresponding described weight coefficient to be calculated, and when result of calculation meets abnormal behaviour threshold value, judge that described moving target is reported to the police as abnormal behaviour; Wherein, label represents result of calculation; What T represented is matrix transpose; Represent the weight coefficient that in described Fusion Features model, each proper vector is corresponding.

In conjunction with the first possible embodiment of first aspect, first aspect to any one possible embodiment in embodiment possible in the 5th, embodiments provide the 6th kind of possible embodiment of first aspect, wherein, described Fusion Features model generates according to following method in advance:

Choose video resource containing abnormal behaviour as positive sample and the video resource containing normal behaviour as negative sample;

Carry out training by support vector machine to described positive sample and described negative sample to calculate, set up and preset Fusion Features model;

Carry out taking turns repetitive exercise to described default Fusion Features model according to described negative sample, and when described default Fusion Features model reaches predetermined threshold value, the default Fusion Features model determining correspondence is the Fusion Features model of practical application more.

In conjunction with first aspect, embodiments provide the 7th kind of possible embodiment of first aspect, wherein, described determine the behavior type of described moving target before, also comprise:

The described moving target of detecting and tracking is classified, determines the characteristic type of described moving target, and when the behavior that the characteristic type of described moving target is behaved being detected, perform the step determining the behavior type of described moving target.

Second aspect, the embodiment of the present invention additionally provides a kind of unusual checking device based on fusion feature, comprising:

Determining unit, for the detecting and tracking result according to moving target in video to be tested, determines the behavior type of described moving target;

Extraction unit, for extracting the various dimensions feature in described moving target; It is multiple that described various dimensions feature comprises in the feature of the following stated moving target: the similarity degree of the intensity of variation of pixel, the arrangement tightness degree of pixel, global shape, two field picture, motion feature, position feature and morphological feature;

Analysis and processing unit, described behavior type characteristic of correspondence Fusion Model for determining according to described determining unit carries out analyzing and processing to the described various dimensions feature that described extraction unit extracts, and judges whether described moving target exists abnormal behaviour according to analysis processing result.

In conjunction with second aspect, embodiments provide the first possible embodiment of second aspect, wherein, described extraction unit comprises:

First computation subunit, for calculating the intensity of variation DisturbRate=FC/FA of the pixel of described moving target in predetermined period; Wherein, FA represents the quantity of all pixels in the foreground area that described predetermined period is corresponding; FC represents the quantity not being in the pixel in described foreground area in foreground area before the certain hour cycle; ;

Second computation subunit, for calculating the arrangement tightness degree CompactRate=CS/CN of the pixel of described moving target in described predetermined period; Wherein, CN represents that in described foreground area, all four neighborhoods are all in the foreground pixel point of described foreground area; CS represents that in CN, all four neighborhoods all exist the foreground pixel point of luminance difference;

3rd computation subunit, for calculating the circularity Ω=P/ (2*sqrt (π A)) of the global shape of described moving target in predetermined period; Wherein, A represents the area of described foreground area; P represents the girth of described foreground area;

4th computation subunit, for calculating the similarity degree AreaSimilarRate=AND/OR of two field picture in described moving target in described predetermined period; Wherein, AND represents that in described foreground area, consecutive frame is all in the number of the pixel of foreground area; Described OR represents the number of the pixel not being in described foreground area in described foreground area in consecutive frame.

In conjunction with second aspect, embodiments provide the embodiment that the second of second aspect is possible, wherein, described extraction unit also comprises:

First extracts subelement, for extracting the motion amplitude of the described moving target in predetermined period in described foreground area; Described motion amplitude comprises: the height of length travel, transversal displacement, current foreground area and the width of current foreground area;

And/or,

5th computation subunit, for calculating the direction of motion OffRate=Offset/Route of described moving target in described predetermined period; Wherein, Offset represents the actual displacement of described moving target, and this displacement comprises transversal displacement and length travel; Route represents the accumulative stroke of all actual displacements in the foreground area in described predetermined period.

In conjunction with second aspect, embodiments provide the third possible embodiment of second aspect, wherein, described extraction unit also comprises:

Second extracts subelement, for extracting the change in displacement value in predetermined period in described moving target between every two two field pictures; Operator unit, for when the described change in displacement value that described second extracts subelement extraction is greater than predetermined threshold value, adds 1 by the variation frequency of the position feature of described moving target; Arrange subelement, for terminating in described predetermined period, the final variation frequency calculated by described operator unit is as the position feature of described moving target.

In conjunction with second aspect, embodiments provide the 4th kind of possible embodiment of second aspect, wherein, described extraction unit also comprises:

6th computation subunit, for calculating the ratio of width to height of the video area at described video place to be tested;

7th computation subunit, for the ratio of width to height of the described video area according to described 6th computation subunit calculating, calculates the ratio of width to height of actual foreground area;

Contrast subunit, contrasts for the ratio of width to height of foreground area of described reality of described 7th computation subunit being calculated and default the ratio of width to height threshold value;

Judgment sub-unit, for judging the actual form that moving target in the foreground area of described reality is corresponding according to the comparing result of described contrast subunit; Described actual form comprises: single region and many people region.

In conjunction with second aspect, embodiments provide the 5th kind of possible embodiment of second aspect, wherein, described analysis and processing unit comprises:

3rd extracts subelement, and for extracting all proper vectors in described various dimensions feature, all proper vectors are expressed as: F=[F1, F2, F3, F4, F5, F6, F7]; Wherein, F represents the set of all proper vectors; Proper vector F1, F2, F3, F4, F5, F6 and F7 are respectively described intensity of variation, described arrangement tightness degree, described global shape, the similarity degree of described two field picture, described motion feature, above-mentioned position feature and described morphological feature;

Obtain subelement, for obtaining all proper vectors and corresponding weight coefficient that described in described Fusion Features model, the 3rd extraction subelement extracts, above-mentioned weight coefficient is expressed as: M=[w1, w2, w3, w4, w5, w6, w7]; Wherein, M represents the set of all weight coefficients; Weight coefficient w1, w2, w3, w4, w5, w6 and w7 are respectively weight coefficient corresponding to described intensity of variation, described arrangement tightness degree, described global shape, the similarity degree of described two field picture, described motion feature, above-mentioned position feature and described morphological feature;

8th computation subunit, for according to formula, calculates described all proper vectors and the corresponding the described 3rd described weight coefficient extracting subelement acquisition, and when result of calculation meets first threshold, judges that described moving target is as abnormal behaviour; Wherein, label represents result of calculation; What T represented is matrix transpose; Represent the weight coefficient that in described Fusion Features model, each proper vector is corresponding.

In conjunction with the first possible embodiment of second aspect, second aspect to any one possible embodiment in the 5th kind of possible embodiment, embodiments provide the 6th kind of possible embodiment of second aspect, wherein, the described unusual checking device based on fusion feature, also comprises:

Choose unit, for choosing video resource containing abnormal behaviour as positive sample and the video resource containing normal behaviour as negative sample;

Training computing unit, carries out training for the positive sample of choosing unit selection to described by support vector machine and described negative sample and calculates, set up default Fusion Features model;

Repetitive exercise computing unit, described negative sample for choosing unit selection described in basis carries out taking turns repetitive exercise to the described default Fusion Features model that described training computing unit is set up more, and when described default Fusion Features model reaches predetermined threshold value, the default Fusion Features model determining correspondence is the Fusion Features model of practical application.

In conjunction with second aspect, embodiments provide the 7th kind of possible embodiment of second aspect, wherein, the described unusual checking device based on fusion feature also comprises:

Taxon, for classifying to the described moving target of detecting and tracking, determines the characteristic type of described moving target;

Described determining unit specifically for, detect described moving target characteristic type behave behavior time, determine the behavior type of described moving target.

A kind of anomaly detection method based on fusion feature that the embodiment of the present invention provides and device, comprising: first according to the detecting and tracking result of moving target in video to be tested, determine the type of moving target, then the various dimensions feature in above-mentioned moving target is extracted, wherein, what above-mentioned various dimensions feature comprised in following characteristics is multiple: the similarity degree of the intensity of variation of moving target, arrangement tightness degree, global shape, two field picture, motion feature, position feature and morphological feature, finally according to the above-mentioned type characteristic of correspondence Fusion Model, analyzing and processing is carried out to various dimensions feature, and judge whether moving target exists abnormal behaviour according to analysis processing result, often in practical application, significant limitation is there is with employing single features of the prior art, easily cause failing to report or report by mistake and comparing of abnormal behaviour, which propose multiple inventive features for abnormal behaviour, can the effectively robustness of boosting algorithm and stability, and according to the best features Fusion Model obtained the learning training of a large amount of abnormal behaviour, analyzing and processing is carried out to above-mentioned various dimensions feature, effectively can avoid over-fitting or the poor fitting problem of algorithm in analytic process, it is applicable to Various Complex application scenarios, save a large amount of time costs and human cost, possesses very high promotional value.

For making above-mentioned purpose of the present invention, feature and advantage become apparent, preferred embodiment cited below particularly, and coordinate appended accompanying drawing, be described in detail below.

Accompanying drawing explanation

In order to be illustrated more clearly in the technical scheme of the embodiment of the present invention, be briefly described to the accompanying drawing used required in embodiment below, be to be understood that, the following drawings illustrate only some embodiment of the present invention, therefore the restriction to scope should be counted as, for those of ordinary skill in the art, under the prerequisite not paying creative work, other relevant accompanying drawings can also be obtained according to these accompanying drawings.

Fig. 1 shows the process flow diagram of a kind of anomaly detection method based on fusion feature that the embodiment of the present invention provides;

Fig. 2 shows another kind that the embodiment of the present invention the provides process flow diagram based on the anomaly detection method of fusion feature;

Fig. 3 shows another kind that the embodiment of the present invention the provides process flow diagram based on the anomaly detection method of fusion feature;

Fig. 4 shows the structural representation of a kind of unusual checking device based on fusion feature that the embodiment of the present invention provides;

Fig. 5 shows a kind of structural representation based on extraction unit in the unusual checking device of fusion feature that the embodiment of the present invention provides;

Fig. 6 shows a kind of structural representation based on analysis and processing unit in the unusual checking device of fusion feature that the embodiment of the present invention provides;

Fig. 7 shows another kind that the embodiment of the present invention the provides structural representation based on the unusual checking device of fusion feature.

Main element symbol description:

11, determining unit; 22, extraction unit; 33, analysis and processing unit; 44, unit is chosen; 55, computing unit is trained; 66, repetitive exercise computing unit; 221, the 6th computation subunit; 222, the 7th computation subunit; 223, contrast subunit; 224, judgment sub-unit; 331, the 3rd subelement is extracted; 332, subelement is obtained; 333, the 8th computation subunit.

Embodiment

Below in conjunction with accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.The assembly of the embodiment of the present invention describing and illustrate in usual accompanying drawing herein can be arranged with various different configuration and design.Therefore, below to the detailed description of the embodiments of the invention provided in the accompanying drawings and the claimed scope of the present invention of not intended to be limiting, but selected embodiment of the present invention is only represented.Based on embodiments of the invention, the every other embodiment that those skilled in the art obtain under the prerequisite not making creative work, all belongs to the scope of protection of the invention.

In order to give full play to supervisory system for the above-mentioned real-time of detection abnormal behaviour and the advantage of initiative, prior art provide a kind of anomaly detection method based on supervisory system, but above-mentioned detection method judges according to the analysis processing result of the single feature of a certain class of moving target (displacement as movement velocity and motion), and adopt above-mentioned single features in practical application, often there is significant limitation, easily cause failing to report or reporting by mistake of abnormal behaviour, and single features can cause the Algorithm robustness of analyzing and processing poor usually, scene requirement is suitable for algorithm also higher, thus also make popularization difficulty higher.

And, average weighted method and linear classification is all adopted to calculate to the process that single features carries out analyzing and processing in above-mentioned detection method, the former is the weighted value mean value of weighted value being set as each single features (displacement as movement velocity and motion) is corresponding, the latter relies on artificial experience to carry out priority level initializing to single features, thus determines the weighted value that each single features is corresponding; And the average weighted method of above-mentioned use and the process of feature weight parameter is set by artificial experience, not only higher to the experience and knowledge level requirement of staff, and easily cause over-fitting or the poor fitting problem of algorithm.

Based on prior art the problems referred to above based on the anomaly detection method of supervisory system, a kind of anomaly detection method based on fusion feature that the embodiment of the present invention provides, often in practical application, significant limitation is there is with employing single features of the prior art, easily cause failing to report or report by mistake and comparing of abnormal behaviour, which propose multiple inventive features for abnormal behaviour, can the effectively robustness of boosting algorithm and stability, and according to the best features Fusion Model obtained the learning training of a large amount of abnormal behaviour, analyzing and processing is carried out to above-mentioned various dimensions feature, effectively can avoid over-fitting or the poor fitting problem of algorithm in analytic process, it is applicable to Various Complex application scenarios, save a large amount of time costs and human cost, possesses very high promotional value.

For the ease of the understanding to the embodiment of the present invention, first brief description is carried out to the supervisory system related in the embodiment of the present invention, above-mentioned supervisory system comprises: monitor camera device, be pre-installed in a lot of public place such as bank, market, station and traffic intersection, for the video image of the above-mentioned public place of Real-time Collection; Surveillance center, comprise server and monitor terminal, be arranged in corresponding government department or relevant safety management department, its video image sent by the above-mentioned monitor camera device of server real-time reception, and analyzing and processing is carried out to this video image, to judge whether have abnormal behaviour in above-mentioned video image, and control warning device warning when abnormal behaviour being detected; It is by the analysis result of the above-mentioned video image of monitor terminal real-time exhibition and above-mentioned video image, with carrying out corresponding operational processes for user's real time inspection.

With reference to the process flow diagram of a kind of anomaly detection method based on fusion feature shown in figure 1, described method comprises the steps:

S102, detecting and tracking result according to moving target in video to be tested, determine the behavior type of described moving target.

In above-mentioned supervisory system, first Surveillance center determines moving target from video to be tested, it can be treated moving targets all in test video in order and carry out detection and tracking process successively, also first according to the extraordinarily thick moving target slightly detecting suspicion abnormal behaviour such as motion state and motion amplitude of moving target, and detection and tracking process can be carried out to these moving targets successively.

Then according to the behavior type of the result determination moving target of above-mentioned detection and tracking process, above-mentioned behavior type can comprise: fight, run, behavior and theft etc. in disaster (as earthquake, fire); And in reality, the above-mentioned behavior type in the embodiment of the present invention is not limited to above-mentioned concrete behavior type.

In addition, moving object detection in the present embodiment and tracking technique are the basis and the prerequisite that realize unusual checking technology, wherein, moving object detection can be able to be divided into the motion detection under static background and the motion under dynamic background to detect according to the relation between target and video camera;

1, the motion under static background is detected: only have target in whole monitor procedure in motion, it mainly comprises following several method.

(1) background subtraction; In whole monitor procedure, need ceaselessly to safeguard one " pure background ", for any frame monitored picture, itself and pure background are carried out difference, thus obtain the moving target that appears in current picture.(2) frame differential method; By the mathematic interpolation between consecutive frame, obtain the method for the information such as the position of moving target, shape, the method is very strong to the adaptive faculty of illumination; Specifically can utilize the mathematic interpolation between adjacent three frames, carry out the detection of moving target.

(3) optical flow method; In space, motion can describe with sports ground; And on a plane of delineation, the motion of object embodies often by the difference of gradation of image distribution in image sequence, thus the sports ground in space is transferred on image be just expressed as optical flow field.Optical flow field reflects the variation tendency of every bit gray scale on image, and its pixel can regarded as with gray scale moves and the instantaneous velocity field produced on the image plane, is also a kind of approximate evaluation to real motion field; In comparatively ideal situation, it can detect the object of self-movement and not need to know in advance any information of scene, very accurately can calculate the speed of moving object, and can be used for the situation of dynamic scene.

2, the motion under dynamic background is detected: in monitor procedure, target and background is all moving or changing; In the applied environment of moving object detection, dynamic background is Comparatively speaking more complicated.According to the forms of motion of camera, following two kinds can be divided into:

(1) camera support is fixed; But camera can rotate along with the motion of The Cloud Terrace, inclination motion of Denging.In addition, camera also can control lens focusing according to remote computer instruction, thus produces distant view and the motion of close shot convergent-divergent.

(2) camera is placed in (such as, in-vehicle camera) on mobile device

For above two kinds of camera motion forms any one for, before carrying out moving object detection, all need to carry out overall motion estimation and compensation according to certain method.Usually, Block Matching Algorithm, Feature Points Matching method etc. can be utilized to carry out momental estimation.Certainly, optical flow method also can be utilized to set up optical flow field model, utilize optical flow equation to solve the movement velocity of image slices vegetarian refreshments.

And motion target tracking is exactly in a continuous videos sequence, in each frame monitored picture, find interested moving target (such as, vehicle, pedestrian, thing (as animal) etc.).Tracking can be roughly divided into following step:

1, effective description of target; The tracing process of target is the same with target detection, needs effectively to describe it, that is, need to extract clarification of objective, thus can express this target; In general, clarification of objective description can be carried out by modes such as the edge of image, profile, shape, texture, region, histogram, moment characteristics, conversion coefficients;

2, similarity measurement calculates; Conventional method has: Euclidean distance, mahalanobis distance, chessboard distance, Weighted distance, similarity coefficient, related coefficient etc.;

3, target area search coupling; If all carry out feature extraction, Similarity measures to all targets occurred in scene, then the calculated amount spent by system cloud gray model is very large.Therefore usually adopt certain mode to estimate the region that moving target may occur at present, thus reduce redundancy, accelerate the speed of target following; Common prediction algorithm has: Kalman filter, particle filter, average drifting etc.

S104, the various dimensions feature extracted in described moving target; It is multiple that described various dimensions feature comprises in the feature of the following stated moving target: the similarity degree of the intensity of variation of pixel, the arrangement tightness degree of pixel, global shape, two field picture, motion feature, position feature and morphological feature.

Concrete, the above-mentioned various dimensions feature extracted in the present embodiment can multiple in above-mentioned feature, and can combination in any between above-mentioned feature; Wherein, above-mentioned motion feature comprises motion amplitude and the direction of motion of the direction of motion of moving target, and above-mentioned position feature comprises the variation frequency of displacement; Above-mentioned two features and other 5 features are all the brand-new characteristic models carrying out targeted Design & reform in the feature of single features (displacement as movement velocity and motion) and the basis of abnormal behaviour feature, the embodiment of the present invention is by the multiple above-mentioned behavioural characteristic of comprehensive utilization, effectively prevent the shortcoming of single features, improve robustness and the stability of algorithm.

S106, according to described behavior type characteristic of correspondence Fusion Model, analyzing and processing is carried out to described various dimensions feature, and judge whether described moving target exists abnormal behaviour according to analysis processing result.

Concrete, when carrying out anomaly analysis according to the upper various dimensions feature extracted to moving target, needing to analyze separately each feature above-mentioned of moving target, when judging that all various dimensions features are abnormal behaviour, determining that moving target is abnormal behaviour.

A kind of anomaly detection method based on fusion feature that the embodiment of the present invention provides, often in practical application, significant limitation is there is with employing single features of the prior art, easily cause failing to report or report by mistake and comparing of abnormal behaviour, which propose multiple inventive features for abnormal behaviour, can the effectively robustness of boosting algorithm and stability, and according to the best features Fusion Model obtained the learning training of a large amount of abnormal behaviour, analyzing and processing is carried out to above-mentioned various dimensions feature, effectively can avoid over-fitting or the poor fitting problem of algorithm in analytic process, it is applicable to Various Complex application scenarios, save a large amount of time costs and human cost, possesses very high promotional value.

Consider that above-mentioned various dimensions feature is the brand-new characteristic model carrying out targeted Design & reform on the basis of the feature of single features (displacement as movement velocity and motion), therefore the respectively leaching process of above-mentioned 7 various dimensions features is being described in detail in the embodiment of the present invention:

The first, for the intensity of variation of the pixel of moving target, above-mentioned intensity of variation also can be expressed as restless degree, and concrete leaching process is as follows:

Following formula DisturbRate=FC/FA is utilized to calculate the intensity of variation of the pixel of described moving target in predetermined period; Wherein, the FA in above-mentioned formula represents the quantity of all pixels in the foreground area that described predetermined period is corresponding; FC represents the quantity not being in the pixel in described foreground area in foreground area before the certain hour cycle.

Concrete, foreground area refers to the zone of action of moving target, and under normal behaviour, this region generally can not change, and this region can change under abnormal behaviour, and therefore corresponding within different time cycles foreground area can be different, consider the problems referred to above, in the embodiment of the present invention, first with same predetermined period, (this predetermined period number can be arranged arbitrarily according to actual needs, be one-period with predetermined period in the present invention, hereafter be referred to as the period 1, and 5 two field pictures are be described in a cycle) be described for example, FA is then the quantity of all pixels in the foreground area within this period 1, and FC represents in all pixels in FA, the certain hour cycle (as one-period) front be not in as described in the quantity of pixel in foreground area, the ratio calculating FC and FA is the intensity of variation of the pixel of moving target or restless degree.

The second, for the arrangement tightness degree of the pixel of moving target, above-mentioned arrangement tightness degree also can be expressed as compactness, and concrete leaching process is as follows:

Utilize following formula CompactRate=CS/CN, calculate the arrangement tightness degree CompactRate=CS/CN of the pixel of described moving target in the described period 1; Wherein, CN represents that in described foreground area, all four neighborhoods are all in the quantity of the foreground pixel point of described foreground area; CS represents that in CN, all four neighborhoods all exist the quantity of the foreground pixel point of luminance difference.

Concrete, contain two unique points in above-mentioned arrangement tightness degree, first be the quantity of pixel, and second is have luminance difference between pixel; The object of carrying out computing by above-mentioned two unique points is the erroneous judgement in order to distinguish the behaviors such as illumination and leaf in video image rock, wherein, when the closeness of the behavior of people is larger, corresponding pixel seems just as a sheet of, and the closeness of the pixel of illumination is more similar with the closeness of the behavior of people, seem it is also a sheet of, but the brightness between the pixel of illumination is the same, these are different from the behavior of people; And the closeness that the leaf of correspondence rocks is also more similar with the closeness of the behavior of people, seem it is also a sheet of, but the less and not all point of the difference between the pixel during leaf rocks is all in foreground area, therefore by erroneous judgement situation that above-mentioned two unique points can avoid above-mentioned illumination and leaf to rock.

Wherein, four neighborhoods in above-mentioned foreground area refer to the circle of four some compositions or the interval of ball of the surrounding in foreground area centered by certain pixel.

Three, for the global shape of moving target, above-mentioned global shape can be expressed as circularity, and its concrete leaching process is as follows:

Following formula Ω=P/2.sqrt (π A) is utilized to calculate the circularity of the global shape of described moving target in predetermined period; Wherein, A represents the area of described foreground area; P represents the girth of described foreground area.

Concrete, the global shape of the video image that the behavior of people is corresponding is circular or oval, and the global shape of the video image of car is rectangle, therefore be to further determine the behavior that the behavior of extracting is behaved by the circularity object calculated, to analyze the abnormal behaviour of people.

Four, for the similarity degree of two field picture in moving target, above-mentioned similarity degree also can be expressed as similarity, and its concrete leaching process is as follows:

According to formula AreaSimilarRate=AND/OR, calculate the similarity degree of two field picture in described moving target in the described period 1; Wherein, AND represents that in described foreground area, consecutive frame is all in the number of the pixel of foreground area; Described OR represents the number of the pixel not being in described foreground area in described foreground area in consecutive frame.

Concrete, in the abnormal behaviour of people, as fought, the similarity of the multiple image in its one-period is very high, this feature can judge in conjunction with the abnormal behaviour of other features to people, make further the algorithm accuracy of the abnormal behaviour of analyst and robustness better.

Wherein, above-mentioned foreground area is still foreground area corresponding to period 1, and AND then represents that in this foreground area, consecutive frame is all in the number of the pixel of foreground area; OR then represents the number of the pixel not being in described foreground area in this foreground area in consecutive frame.

Five, for the motion feature of moving target, above-mentioned motion feature comprises motion amplitude and direction of motion, and above-mentioned direction of motion can be expressed as one-way; Concrete, the concrete leaching process of above-mentioned motion feature is as follows:

And/or, according to formula OffRate=Offset/Route, calculate the direction of motion of described moving target in described predetermined period; Wherein, Offset represents the actual displacement of described moving target, and this displacement comprises transversal displacement and length travel; Route represents the accumulative stroke of all actual displacements in the foreground area in described predetermined period.

Concrete, above-mentioned motion feature can comprise motion amplitude and direction of motion two features, first extracts the motion amplitude features such as the length travel of moving target in foreground area, transversal displacement, the height of current foreground area and the width of current foreground area; In addition, can also according to above-mentioned motion amplitude feature calculation direction of motion feature (i.e. one-way);

Wherein, the period 1, molecule Offset was the current displacement point deducting the first frame from the current displacement point of the 5th frame for 5 two field pictures, was the displacement of the actual generation of moving target; Denominator Route is the cumulative of the absolute value of the intrinsic displacement of every frame within 5 frames; Then when the OffRate obtained is 1, be then judged to be one-way, therefore corresponding one-way OffRate of running facing one direction is then close to 1.

Wherein, laterally ColOffRate=ColOffset/ColRoute quantizes with the formula, and calculate horizontal one-way: wherein, ColOffset is the horizontal actual displacement of moving target, and ColRoute is that the horizontal interframe being in foreground area in the period 1 adds up stroke; And horizontal one-way ColOffRate is close to the horizontal one-way movement of 1 expression.

In like manner, longitudinally RowOffRate=RowOffset/RowRoute quantizes with the formula, and calculate longitudinal one-way: wherein, RowOffset is longitudinal actual displacement of moving target, and RowRoute is that the longitudinal interframe being in foreground area in the period 1 adds up stroke; And longitudinal one-way RowOffRate is close to the longitudinal one-way movement of 1 expression.

Six, for the position feature of moving target, the concrete leaching process of above-mentioned motion feature is as follows:

Concrete, as long as the displacement change of moving target all calculates shift in position in reality, but in the present embodiment, be previously provided with a movement threshold, when shift in position exceedes this movement threshold, just be judged to be shift in position, the object of movement threshold is set in order to distinguish the object as time display screen etc., its position remains constant but word on it changes always, if the time is ceaselessly dynamic.

Wherein, the concrete account form of the above-mentioned variation frequency is as follows: still for the period 1, subtraction is done in position between every two adjacent moving targets in foreground area in period 1, as difference is greater than movement threshold 5, then judge shift in position, the variation frequency adds 1, until after the period 1 terminates, the final variation frequency is as the position feature of described moving target.

It should be noted that, above-mentioned predetermined threshold value can be arranged arbitrarily as required, and the present invention does not do concrete restriction to its concrete numerical value.

Seven, for the morphological feature of moving target, the concrete leaching process following steps of above-mentioned morphological feature, with reference to figure 2:

S202, calculate the ratio of width to height of the video area at described video place to be tested.

S204, the ratio of width to height according to described video area, calculate the ratio of width to height of actual foreground area.

S206, the ratio of width to height of the foreground area of described reality and default the ratio of width to height threshold value to be contrasted, and judge the actual form that moving target in the foreground area of described reality is corresponding according to comparing result; Described actual form comprises: single form and many people form.

Concrete, single form is different with the ratio of width to height that many people form forms foreground target (moving target namely in foreground area) region, and this ratio of width to height with video itself is relevant; As compressed at video height itself, single the ratio of width to height will be made close to the ratio of width to height of many people; As normal video, single the ratio of width to height is 1:4, is 1:2 after being compressed; And normal many people the ratio of width to height is also 1:2, so time also need to judge single form and many people form according to the ratio of width to height of video itself.

Concrete computation process is namely: the ratio of width to height first calculating video area, then in the ratio of width to height calculating actual foreground area, the ratio of width to height in actual foreground region and default form threshold value are compared, as being be judged to be single form being less than default form threshold value, be judged to be many people form being greater than default form threshold value.

It should be noted that, above-mentioned default form threshold value can be arranged arbitrarily as required, and the present invention does not do concrete restriction to its concrete numerical value.

Above-mentioned 7 multidimensional characteristic characteristic of correspondence values such as restless degree, compactness, circularity, similarity, motion feature, position feature and shape facility are represented with F1 ~ F7 respectively in the embodiment of the present invention.

After being extracted above-mentioned 7 multidimensional characteristics, housing choice behavior type characteristic of correspondence Fusion Model, and according to the above-mentioned Fusion Features model of the correspondence selected, analyzing and processing is carried out to above-mentioned various dimensions feature, and judge whether described moving target exists abnormal behaviour according to analysis processing result, wherein, above-mentioned analyzing and processing and judge that the specific implementation of abnormal behaviour is as follows:

First extract all proper vectors in described various dimensions feature, above-mentioned all proper vectors can be expressed as: F=[F1, F2, F3, F4, F5, F6, F7]; Wherein, F represents the set of all proper vectors; Proper vector F1, F2, F3, F4, F5, F6 and F7 are respectively described intensity of variation, described arrangement tightness degree, described global shape, the similarity degree of described two field picture, described motion feature, above-mentioned position feature and described morphological feature;

According to formula described all proper vectors and corresponding described weight coefficient are calculated, and when result of calculation meets first threshold, judges that described moving target is as abnormal behaviour; Wherein, label represents result of calculation; What T represented is matrix transpose; Wi represents the weight coefficient that in described Fusion Features model, each proper vector is corresponding.

Concrete, the behavior type characteristic of correspondence Fusion Model of selection can be expressed as: M=[w1, w2, w3, w4, w5, w6, w7], and above-mentioned 7 weight coefficient w1 ~ w7 of above-mentioned 7 proper vector F1 ~ F7 and correspondence are all updated to formula in, the result that label is corresponding can be calculated, and when label=-1 (i.e. abnormal behaviour threshold value), then judge that moving target is normal condition; When label=1 (i.e. normal behaviour threshold value), then detect in moving target have abnormal behaviour, and report to the police, safeguard in order to remind staff.

Consider in prior art the over-fitting or the poor fitting problem that use average weighted method and easily caused algorithm by the process that artificial experience arranges feature weight parameter, the embodiment of the present invention adopts support vector machine, a large amount of training sample carries out Training, obtain theoretical best features Fusion Model, the method can effectively be avoided arranging by experience the algorithm over-fitting or poor fitting problem that feature weight parameter causes.With reference to figure 3, the Fusion Features model in the embodiment of the present invention generates according to following method in advance, specifically comprises the steps:

S302, choose video resource containing abnormal behaviour as positive sample and the video resource containing normal behaviour as negative sample.

Concrete, choose the video segment containing abnormal behaviour, extract the above-mentioned 7 class multidimensional characteristics of the moving target between the abnormal behaviour emergence period, and as the positive sample of training sample set; Choose containing other normal video segments, random above-mentioned 7 category features extracting the moving target of any time, and as the negative sample of training sample set.

S304, by support vector machine to described positive sample and described negative sample carry out training calculate, set up preset Fusion Features model.

Concrete, adopt support vector machine to be that sorter is trained above-mentioned positive sample and negative sample, and obtain Fusion Features model by many wheel repetitive exercise; Wherein, different according to the Fusion Features model that different behavior types trains, as corresponding a kind of Fusion Features model of fighting, and corresponding another kind of Fusion Features model of running; And the Fusion Features model of different behavior types, the weighted value (or claiming weight coefficient) corresponding to it proper vector comprised is different.

S306, carry out taking turns repetitive exercise to described default Fusion Features model according to described negative sample, and when described default Fusion Features model reaches predetermined threshold value, the default Fusion Features model determining correspondence is the Fusion Features model of practical application more.

Consider the integrity problem of training the above-mentioned Fusion Features model obtained, carrying out taking turns repetitive exercise to the above-mentioned Fusion Features model established according to negative sample more, in the characteristic model that above-mentioned training obtains as constantly substituted into by negative sample, if judged result all matches with actual abnormal behaviour in the first preset times (as 100 times), then think that above-mentioned characteristic model is reliable, and it can be used as the Fusion Features model of practical application; Do not mate with time anomaly behavior (namely all exporting as normal behaviour) if judged result is in preset times (as 20 times), then think that above-mentioned characteristic model is unreliable, now, when the judged result exported is not mated with actual result, support vector machine all repeatedly adjusts the weighted value stating each proper vector in above-mentioned characteristic model, until obtain the result all matched with actual abnormal behaviour in the first preset times (as 100 times), then it can be used as the Fusion Features model of practical application.

In addition, anomaly detection method in the embodiment of the present invention is mainly for the abnormal behaviour of people, in order to more moving target be determined for people, in step 102 " before determining the behavior type of described moving target ", also comprise: the described moving target of detecting and tracking is classified, determines the characteristic type of described moving target, then, when the behavior that the characteristic type of described moving target is behaved being detected, carrying out the step of the behavior type determining described moving target.

In the embodiment of the present invention, first can according to unique points such as the displacements of the movement velocity of moving target, moving target size and motion object, the described moving target of detecting and tracking is classified, judge that it belongs in people's behavior on earth, or during garage is, or thing (as illumination and leaf etc.) behavior.And when detecting that above-mentioned moving target is the behavior of people, carrying out the step of the behavior type determining described moving target.

With reference to figure 4, the embodiment of the present invention also provides a kind of unusual checking device based on fusion feature, comprising:

Determining unit 11, for the detecting and tracking result according to moving target in video to be tested, determines the behavior type of described moving target;

Extraction unit 22, for extracting the various dimensions feature in described moving target; It is multiple that described various dimensions feature comprises in the feature of the following stated moving target: the similarity degree of the intensity of variation of pixel, the arrangement tightness degree of pixel, global shape, two field picture, motion feature, position feature and morphological feature;

Analysis and processing unit 33, described behavior type characteristic of correspondence Fusion Model for determining according to described determining unit carries out analyzing and processing to the described various dimensions feature that described extraction unit extracts, and judges whether described moving target exists abnormal behaviour according to analysis processing result.

A kind of unusual checking device based on fusion feature that the embodiment of the present invention provides, often in practical application, significant limitation is there is with employing single features of the prior art, easily cause failing to report or report by mistake and comparing of abnormal behaviour, which propose multiple inventive features for abnormal behaviour, can the effectively robustness of boosting algorithm and stability, and according to the best features Fusion Model obtained the learning training of a large amount of abnormal behaviour, analyzing and processing is carried out to above-mentioned various dimensions feature, effectively can avoid over-fitting or the poor fitting problem of algorithm in analytic process, it is applicable to Various Complex application scenarios, save a large amount of time costs and human cost, possesses very high promotional value.

Consider that above-mentioned various dimensions feature is the brand-new characteristic model carrying out targeted Design & reform on the basis of the feature of single features (displacement as movement velocity and motion), therefore also needing in the embodiment of the present invention respectively the carrying out of above-mentioned 7 various dimensions features is extracted, corresponding extraction unit 22 comprises:

The first, for the intensity of variation of the pixel of moving target, above-mentioned intensity of variation also can represent restless degree;

The second, for the arrangement tightness degree of the pixel of moving target, above-mentioned arrangement tightness degree also can represent compactness;

Three, for the global shape of moving target, above-mentioned global shape can represent the circularity calculating moving target;

Four, for the similarity degree of two field picture in moving target, above-mentioned similarity degree also can be expressed as similarity;

Concrete, the leaching process of above-mentioned restless degree, compactness, circularity and similarity is as follows: described extraction unit 22 comprises:

First computation subunit, for calculating the intensity of variation DisturbRate=FC/FA of the pixel of described moving target in predetermined period; Wherein, FA represents the quantity of all pixels in the foreground area that described predetermined period is corresponding; FC represents the quantity not being in the pixel in described foreground area in foreground area before the certain hour cycle;

Five, for the motion feature of moving target, above-mentioned motion feature comprises: motion amplitude and direction of motion, and above-mentioned direction of motion can be expressed as one-way; Concrete, the concrete leaching process of above-mentioned motion feature is as follows: described extraction unit 22 also comprises:

And/or,

Six, for the position feature of moving target, the concrete leaching process of above-mentioned motion feature is as follows: described extraction unit 22 also comprises:

Seven, for the morphological feature of moving target, the concrete leaching process of above-mentioned morphological feature is as follows: with reference to figure 5, and described extraction unit 22 also comprises:

6th computation subunit 221, for calculating the ratio of width to height of the video area at described video place to be tested;

7th computation subunit 222, for the ratio of width to height of the described video area according to described 6th computation subunit 221 calculating, calculates the ratio of width to height of actual foreground area;

Contrast subunit 223, contrasts for the ratio of width to height of the foreground area of described reality of described 7th computation subunit 222 being calculated and default the ratio of width to height threshold value;

Judgment sub-unit 224, the actual form that the moving target for judging in the foreground area of described reality according to the comparing result of described contrast subunit 223 is corresponding; Described actual form comprises: single region and many people region.

Wherein, the above-mentioned restless degree of said extracted, compactness, circularity, similarity, motion feature, above-mentioned 7 multidimensional characteristic characteristic of correspondence values such as position feature and shape facility represent with F1 ~ F7 respectively.

After being extracted above-mentioned 7 multidimensional characteristics, housing choice behavior type characteristic of correspondence Fusion Model, and according to the above-mentioned Fusion Features model of the correspondence selected, analyzing and processing is carried out to above-mentioned various dimensions feature, and judge whether described moving target exists abnormal behaviour according to analysis processing result, wherein, the specific implementation of corresponding analysis and processing unit is as follows:

Further, with reference to figure 6, above-mentioned based in the unusual checking device of fusion feature, described analysis and processing unit 33 comprises:

3rd extracts subelement 331, and for extracting all proper vectors in described various dimensions feature, all proper vectors are expressed as: F=[F1, F2, F3, F4, F5, F6, F7]; Wherein, F represents the set of all proper vectors; Proper vector F1, F2, F3, F4, F5, F6 and F7 are respectively described intensity of variation, described arrangement tightness degree, described global shape, the similarity degree of described two field picture, described motion feature, above-mentioned position feature and described morphological feature;

Obtain subelement 332, for obtaining all proper vectors and corresponding weight coefficient that described in described Fusion Features model, the 3rd extraction subelement extracts, above-mentioned weight coefficient is expressed as: M=[w1, w2, w3, w4, w5, w6, w7]; Wherein, M represents the set of all weight coefficients; Weight coefficient w1, w2, w3, w4, w5, w6 and w7 are respectively weight coefficient corresponding to described intensity of variation, described arrangement tightness degree, described global shape, the similarity degree of described two field picture, described motion feature, above-mentioned position feature and described morphological feature;

8th computation subunit 333, for according to formula extract to described all proper vectors and the corresponding the described 3rd the described weight coefficient that subelement obtains to calculate, and when result of calculation meets abnormal behaviour threshold value, judge that described moving target is reported to the police as abnormal behaviour; Wherein, label represents result of calculation; What T represented is matrix transpose; w _irepresent the weight coefficient that in described Fusion Features model, each proper vector is corresponding.

Consider in prior art the over-fitting or the poor fitting problem that use average weighted method and easily caused algorithm by the process that artificial experience arranges feature weight parameter, the embodiment of the present invention is by setting up Fusion Features model in advance with lower device.Concrete, with reference to figure 7, the above-mentioned unusual checking device based on fusion feature also comprises:

Choose unit 44, for choosing video resource containing abnormal behaviour as positive sample and the video resource containing normal behaviour as negative sample;

Training computing unit 55, carries out training for the positive sample of choosing unit selection to described by support vector machine and described negative sample and calculates, set up default Fusion Features model;

Repetitive exercise computing unit 66, described negative sample for choosing unit selection described in basis carries out taking turns repetitive exercise to the described default Fusion Features model that described training computing unit is set up more, and when described default Fusion Features model reaches predetermined threshold value, the default Fusion Features model determining correspondence is the Fusion Features model of practical application.

In addition, anomaly detection method in the embodiment of the present invention is mainly for the abnormal behaviour of people, in order to more moving target be determined for people, above-mentioned based in the unusual checking device of fusion feature, also comprise: taxon, for classifying to the described moving target of detecting and tracking, determine the characteristic type of described moving target;

Described determining unit 11 specifically for, detect described moving target characteristic type behave behavior time, determine the behavior type of described moving target.

The carrying out that the embodiment of the present invention provides is based on the computer program of the anomaly detection method of fusion feature, comprise the computer-readable recording medium storing program code, the instruction that described program code comprises can be used for performing the method described in previous methods embodiment, specific implementation see embodiment of the method, can not repeat them here.

Those skilled in the art can be well understood to, and for convenience and simplicity of description, the specific works process of the system of foregoing description, device and unit, with reference to the corresponding process in preceding method embodiment, can not repeat them here.

In several embodiments that the application provides, should be understood that disclosed system, apparatus and method can realize by another way.Device embodiment described above is only schematic, such as, the division of described unit, be only a kind of logic function to divide, actual can have other dividing mode when realizing, again such as, multiple unit or assembly can in conjunction with or another system can be integrated into, or some features can be ignored, or do not perform.Another point, shown or discussed coupling each other or direct-coupling or communication connection can be by some communication interfaces, and the indirect coupling of device or unit or communication connection can be electrical, machinery or other form.

The described unit illustrated as separating component or can may not be and physically separates, and the parts as unit display can be or may not be physical location, namely can be positioned at a place, or also can be distributed in multiple network element.Some or all of unit wherein can be selected according to the actual needs to realize the object of the present embodiment scheme.

In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, also can be that the independent physics of unit exists, also can two or more unit in a unit integrated.

If described function using the form of SFU software functional unit realize and as independently production marketing or use time, can be stored in a computer read/write memory medium.Based on such understanding, the part of the part that technical scheme of the present invention contributes to prior art in essence in other words or this technical scheme can embody with the form of software product, this computer software product is stored in a storage medium, comprising some instructions in order to make a computer equipment (can be personal computer, server, or the network equipment etc.) perform all or part of step of method described in each embodiment of the present invention.And aforesaid storage medium comprises: USB flash disk, portable hard drive, ROM (read-only memory) (ROM, Read-OnlyMemory), random access memory (RAM, RandomAccessMemory), magnetic disc or CD etc. various can be program code stored medium.

The above; be only the specific embodiment of the present invention, but protection scope of the present invention is not limited thereto, is anyly familiar with those skilled in the art in the technical scope that the present invention discloses; change can be expected easily or replace, all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of described claim.

Claims

1. based on an anomaly detection method for fusion feature, it is characterized in that, comprising:

2. the anomaly detection method based on fusion feature according to claim 1, is characterized in that, the various dimensions feature extracted in described moving target comprises:

Calculate the intensity of variation DisturbRate=FC/FA of the pixel of described moving target in predetermined period; Wherein, FA represents the quantity of all pixels in the foreground area that described predetermined period is corresponding; FC represents the quantity not being in the pixel in described foreground area in foreground area before the certain hour cycle;

3. the anomaly detection method based on fusion feature according to claim 1, is characterized in that, the various dimensions feature extracted in described moving target comprises:

And/or,

4. the anomaly detection method based on fusion feature according to claim 1, is characterized in that, the various dimensions feature extracted in described moving target comprises:

5. the anomaly detection method based on fusion feature according to claim 1, is characterized in that, the various dimensions feature extracted in described moving target comprises:

6. the anomaly detection method based on fusion feature according to claim 1, it is characterized in that, according to described behavior type characteristic of correspondence Fusion Model, analyzing and processing is carried out to described various dimensions feature, and judges whether described moving target exists abnormal behaviour and comprise according to analysis processing result:

According to formula described all proper vectors and corresponding described weight coefficient to be calculated, and when result of calculation meets abnormal behaviour threshold value, judge that described moving target is reported to the police as abnormal behaviour; Wherein, label represents result of calculation; What T represented is matrix transpose; w _irepresent the weight coefficient that in described Fusion Features model, each proper vector is corresponding.

7. the anomaly detection method based on fusion feature according to claim 1-6 any one, is characterized in that, described Fusion Features model generates according to following method in advance:

8. the anomaly detection method based on fusion feature according to claim 1, is characterized in that, described determine the behavior type of described moving target before, also comprise:

9., based on a unusual checking device for fusion feature, it is characterized in that, comprising:

10. the unusual checking device based on fusion feature according to claim 9, it is characterized in that, described extraction unit comprises:

The 11. unusual checking devices based on fusion feature according to claim 9, it is characterized in that, described extraction unit also comprises:

And/or,

The 12. unusual checking devices based on fusion feature according to claim 9, it is characterized in that, described extraction unit also comprises:

The 13. unusual checking devices based on fusion feature according to claim 9, it is characterized in that, described extraction unit also comprises:

The 14. unusual checking devices based on fusion feature according to claim 9, it is characterized in that, described analysis and processing unit comprises:

8th computation subunit, for according to formula described all proper vectors and the corresponding the described 3rd described weight coefficient extracting subelement acquisition are calculated, and when result of calculation meets first threshold, judges that described moving target is as abnormal behaviour; Wherein, label represents result of calculation; What T represented is matrix transpose; w _irepresent the weight coefficient that in described Fusion Features model, each proper vector is corresponding.

15. unusual checking devices based on fusion feature according to claim 9-14 any one, is characterized in that, also comprise:

The 16. unusual checking devices based on fusion feature according to claim 9, is characterized in that, also comprise: