CN106951827A - A kind of global abnormal behavioral value method based on target kinetic characteristic - Google Patents
A kind of global abnormal behavioral value method based on target kinetic characteristic Download PDFInfo
- Publication number
- CN106951827A CN106951827A CN201710092206.5A CN201710092206A CN106951827A CN 106951827 A CN106951827 A CN 106951827A CN 201710092206 A CN201710092206 A CN 201710092206A CN 106951827 A CN106951827 A CN 106951827A
- Authority
- CN
- China
- Prior art keywords
- motion vector
- abnormal
- kinetic characteristic
- target
- behavioral value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of global abnormal behavioral value method based on target kinetic characteristic, belong to the technical field of video behavioral value, comprise the following steps:Extract the kinetic characteristic of moving target in video;Moving target is carried out based on energy, based on dispersion and based on the dynamic (dynamical) abnormal behaviour model learning of Lagrangian particle and establishment respectively according to extracted kinetic characteristic;According to the global abnormal behavioral value based on energy, based on dispersion and based on the dynamic (dynamical) abnormal behaviour model progress moving target of Lagrangian particle established, determine abnormal area and be marked.Inherent characteristicses of the present invention according to video abnormal behaviour, can effectively improve the accuracy of global abnormal behavioral value and the efficiency of detection, carry out unusual checking using three models, further increase accuracy of detection.
Description
Technical field
The present invention relates to a kind of global abnormal behavioral value method based on target kinetic characteristic, belong to video abnormal behaviour
The technical field of detection.
Background technology
Video unusual checking refers to that the abnormal behaviour occurred under video monitoring scene can be automatically analyzed, and it can be sent out
Go out alarm signal.For example:Crowd hides everywhere suddenly in crowd scatters everywhere suddenly on spacious lawn behavior, corridor
The behavior that crowd flees in all directions suddenly in behavior, open square, intelligent monitor system can in time detect and carry out alarm.It is false
It is located on the basis of one group of normal training video sample of acquisition, then determines whether test video includes abnormal behaviour.
The research work of abnormal in early stage behavioral value is focused primarily upon describes human body behavior, such as base using simple set model
In two-dimensional silhouette model, three-dimensional cylinder body Model etc.;In addition to static geometric model, researcher, which also attempts to utilize, describes human body fortune
Dynamic some features are modeled, and such as shape, angle, position, movement velocity, the direction of motion, movement locus feature are entered
Row behavior description and differentiation, and using the subspace method including PCA, independent component analysis method etc. to carrying
The feature taken carries out dimensionality reduction and screening, so as to carry out behavioural analysis.The existing invention for unusual checking, exists and fails very
Just understanding the inherent characteristicses of abnormal behaviour, thus existing unusual checking model can not abnormal reaction behavior completely sheet
Matter, so as to cause the accuracy of detection that is obtained according to existing unusual checking model and not up to ideal effect.
The content of the invention
The technical problems to be solved by the invention are to overcome the deficiencies in the prior art to move spy based on target there is provided one kind
Property global abnormal behavioral value method, solve of overall importance not detecting in existing video abnormality detection method, reduction detection
The problem of effect.
It is of the invention specific using following technical scheme solution above-mentioned technical problem:
A kind of global abnormal behavioral value method based on target kinetic characteristic, comprises the following steps:
Step A, the kinetic characteristic for extracting moving target in video;
Step B, extract according to step A kinetic characteristic moving target is carried out based on energy, based on dispersion and base respectively
In the dynamic (dynamical) abnormal behaviour model learning of Lagrangian particle and establishment;
Step C, according to establish based on energy, based on dispersion and based on the dynamic (dynamical) abnormal behaviour mould of Lagrangian particle
Type carries out the global abnormal behavioral value of moving target, determines abnormal area and is marked.
Further, as a preferred technical solution of the present invention:The kinetic characteristic bag of moving target in the step A
Include:Space-time characteristic, motion vector characteristic, motion vector direction histogram, motion vector intensity, motion vector mesh scale.
Further, as a preferred technical solution of the present invention:Abnormal behaviour mould based on energy in the step B
Type learns, and is specially:
Determine the degree of contact of two particle potential energy in solid:
Wherein, U (r) represents the potential energy between two particles, and r represents their Euclidean distance, and a, b, m, n represents constant,
And m>n;
The interaction force between two particle is calculated in conjunction with power:
Kinetic characteristic is extracted with reference to step A, each direction of motion vector intensity is modified:
Wherein, riFor foreground area and the speed of background area, i is the quantity of video frame image, and j is the number of moving mass
Amount, k is the quantity of motion vector.W (r) represents connection weight of the distance for r two particles,For in i-th of frame of video
The direction of k-th of motion vector intensity of j moving mass,For improved motion vector intensity.
The threshold value in each direction of the motion vector intensityDepend onMaximum:
Further, as a preferred technical solution of the present invention:Abnormal behaviour mould based on dispersion in the step B
Type learns and established, and is specially:
Determine that the threshold value in each point direction of motion vector intensity is depended onMaximum:
According to the definition of entropy and combination entropy, determine interactive information be the definition I (X, Y) based on entropy=H (X)+H (Y)-H (X,
Y), wherein, H (X) and H (Y) are X and the Shannon entropy of Y-direction respectively;H (X, Y) is combination entropy;
The interactive information of normalized combination entropy is represented using normalized mutual information, it is defined as follows:
Wherein, NMI (X, Y) has fixed lower bound 0, the fixed upper bound 1.
Further, as a preferred technical solution of the present invention:Lagrangian particle power is based in the step B
Abnormal behaviour model learning, including:The calculating of normalized direction histogram interactive information:
Wherein, γ ∈ [0,1] are fusion parameters, and σ represents standard deviation;NMI (X, Y) is the interactive information of combination entropy;
For the threshold value in each direction.
Further, as a preferred technical solution of the present invention:The global different of moving target is carried out in the step C
Normal behavioral value includes:Moving target is pre-processed, the pretreatment includes filtering and smoothing processing
The present invention uses above-mentioned technical proposal, can produce following technique effect:
A kind of global abnormal behavioral value method based on target kinetic characteristic that the present invention is provided, extracts tracking mesh first
Target kinetic characteristic;Then, moving target is carried out based on energy, based on dispersion and based on Lagrangian particle dynamics respectively
Abnormal behaviour model learning;Finally, the global abnormal behavioral value of moving target is carried out according to learning outcome, exceptions area is determined
Domain is simultaneously effectively marked.Global abnormal behavioral value method of the invention based on target kinetic characteristic can effectively improve the overall situation
The accuracy of unusual checking and the efficiency of detection.Compared to existing invention, the advantage of the invention is that according to abnormal row
For inherent characteristicses, the model based on energy, the model based on dispersion are used successively, based on the dynamic (dynamical) mould of Lagrangian particle
Type carries out unusual checking, so as to further increase the precision of unusual checking.
Brief description of the drawings
Fig. 1 is the schematic flow sheet of the global abnormal behavioral value method of the invention based on target kinetic characteristic.
The schematic diagram of normal behaviours of the Fig. 2 (a) for the present invention under the scene of hall, Fig. 2 (b) is the present invention in hall scene
The schematic diagram of lower abnormal behaviour.
The schematic diagram without mark of normal behaviours of the Fig. 3 (a) for the present invention under the scene of corridor, Fig. 3 (b) and (c) are this
The schematic diagram that invention is marked in abnormal behaviour.
Embodiment
Embodiments of the present invention are described with reference to Figure of description.
As shown in figure 1, the present invention proposes a kind of global abnormal behavioral value method based on target kinetic characteristic, this method
Specifically include following steps:
Step A, the kinetic characteristic for extracting moving target in video;
Step B, extract according to step A kinetic characteristic moving target is carried out based on energy, based on dispersion and base respectively
In the dynamic (dynamical) abnormal behaviour model learning of Lagrangian particle and establishment;
Step C, according to establish based on energy, based on dispersion and based on the dynamic (dynamical) abnormal behaviour mould of Lagrangian particle
Type carries out the global abnormal behavioral value of moving target, determines abnormal area and is marked.
Specifically, it is described during unusual checking in the step A, it is from motion video the step of most critical
It is middle to extract most important, most rational movable information.The Optic flow information in speed and direction is included for extraction herein.Preferably move mesh
Target kinetic characteristic includes:Space-time characteristic, motion vector characteristic, motion vector direction histogram, motion vector intensity, motion arrow
Measure mesh scale.
Step A1, video frequency motion target space-time characteristic extraction process it is as follows:
(1) space characteristics.Consider computational efficiency and logic reasonability, the spacial characteristics model definition that the present invention is used is such as
Under:
S (x, y)=| | F-1(ei·P(F(I(x,y))))||
Wherein, S (x, y) representation space feature, I (x, y) represents the frame of video given in video, F and F-1Fu is represented respectively
In leaf transformation and inverse Fourier transform, P () represent phase spectrum.
(2) temporal characteristics.First, the difference of two continuous videos interframe is calculated, to obtain global change;Then, extract
The notable feature of different video interframe, to obtain dynamic change.Computing formula is as follows:
Wherein, d (x, y) represents different picture frames, and i and i-2 represent the space-time characteristic of frame of video.Calculated not every a frame
With the temporal characteristics of frame of video, disturbed to avoid introducing in Video coding and video decoding process.
Step A2, the motion vector characteristic extraction process it is as follows:
Motion vector is represented by a line segment with specific direction, this line segment connection initial point p and terminal q.When
The coordinate of initial point isMotion vectorIt is the tracking characteristics by a series of two consecutive frames i and i-1.By initial positionExercise intensityThe direction of motionAnd movement velocityComposition, formula is as follows:
Wherein i is frame number, and j is block number, and k is the quantity of motion vector.
Exercise intensityCalculated by below equation:
In addition, movement velocityCalculated by below equation:
Wherein m represents the quantity of motion vector in jth block in the i-th frame, and ψ represents the center of space-time characteristic at (x, y).
Motion vectorDirectionIt is to be determined by the coordinate of consecutive frame initialization points.Group movement main movement direction
Calculating need consider each direction of motion characteristic point intensive percentage.When the percentage is more than given threshold, then it is assumed that
It is chaotic in this direction.
To improve computational efficiency, divide the image into as m × v blocks.The particle energy of the i-th two field picture is by the frame figure in the present invention
The prospect I of picturefWith the ratio r of backgroundiDetermine:
Wherein IwAnd IhIt is the width and height of background image respectively.
The histogrammic extraction process of step A3, motion vector direction is as follows:
Motion vector direction histogram:36 containers of the invention that the π of direction 2 is divided into direction histogram, and production
Histogram is used for distribution and the direction for representing target motion.2 π are such as divided into 8 direction histograms, then per part orientation angle
Between [α, α+45], wherein α ∈ { 0,45,90,135,180,225,270,315 } can be expressed from the next:
Wherein, h is the quantity of direction histogram container, and the present invention is taken as the quantity that 36, N is quantification gradation.
, therefore, can be by adjustment side because the density percent ratio of each direction character point is presented as in the direction of motion vector
To window wiTo distinguish main and secondary crowd movement direction.By based on dominance direction window analysis per frame direction histogram,
Then the opposite direction of motion vector can be detected by Direction Probability distribution.To each direction window wiSize can be under
Threshold value shown in formulaJudged:
Wherein,Represent the sum of vector.
By finding histogrammic Main way window wi, x-axis intercept by intersecting acquisition with figure,By following
Formula is calculated:
When vector quantities exceed threshold valueWhen, number of containers is expressed as b histogram X:
X∈{x1,x2,...,xa,...,xn, 0≤a < N (b), a ∈ N
Therefore, main orientation window wiBe calculated as follows:
Step A4, motion vector intensity extraction process it is as follows:
Under normal circumstances, the motion of static nature is less than minimum strength.By contrast, the intensity of noise characteristic considerably beyond
General threshold value.Exercise intensity is set in experiment from minimum every frame ε (=1.0e-6) pixel is to maximum every framePicture
Element, the maximum and minimum value are obtained by testing.Such as target speed faster, then the more features of motion vector will lose
Lose.Therefore, ideally, the motion vector of a target is expressed as 4 × 4 × 3.In addition, when mesh point is 3 × 3, the mesh
Scale is 9 × 9 pixels, and moved with 45 pixel per second.The motion vector quantity in each direction can be calculated by below equation
Obtain:
Wherein,Represent t motion vector sum.
It can be estimated per the distribution of frame motion vector Direction Probability by direction histogram:
Wherein,Represent t motion vector sum.
Motion vector sum can be reduced as reverse motions target increases, it means that distribution of movement must consider that motion is strong
Correlation between degree and the direction of motion is calculated, therefore, and the present invention uses the intensity and speed of following formula calculation of motion vectors:
Step A5, motion vector mesh scale extraction process it is as follows:
Direction of motion histogram Z normalized form is:
As described in motion vector direction histogram, when size of mesh opening is minimum, each main orientation window wiTake maximum.
If size of mesh opening is maximum, it will be reduced in the motion vector quantity of congested area;On the other hand, in rarefaction
The motion vector quantity in domain can't change.As video sequence motion feature quantity keeps constant and consecutive frame feature quantity phase
To stable, then these features are optimized using the mask in the crowd of expression region.If gridding size is minimum, motion
The variance of intensity is also minimum.
And, the step B according to step A extract kinetic characteristic to moving target respectively carry out based on energy, be based on
Dispersion and based on the dynamic (dynamical) abnormal behaviour model learning of Lagrangian particle and establishment.It is specific as follows:
Step B1, the abnormal behaviour model learning based on energy and establishment.
The potential energy of two particles in solid, can be represented with the degree of contact shown in following formula:
Wherein U (r) represents the potential energy between two particles, and r represents their Euclidean distance, a, b, m, n (usual m>N) table
Show constant.Section 1 in formula represents the potential energy field of mutual exclusion, and Section 2 represents the potential energy field attracted.When r very littles, two grains
Son is in mutual exclusion state;When r is larger, two particles are in attraction state.
The interaction force between two particle is calculated in conjunction with power.If apart from each other between two particle, interparticle combination
Power also can be weaker.Here, adhesion is expressed as U (r) negative derivative:
Wherein parameter a, b, m, n are respectively set to 1,1,3 and 2.When r isDuring individual pixel, function takes minimum value.With Increase, f (r) absolute value constantly reduces.
Because distance is much larger than between two moving targets in practiceSo as to cause f (r) variation tendency to meet need
Ask, therefore, replace original function using f (r) derivative, and further standardize:
Wherein w (r) represents the connection weight of the two particle at a distance of r.
Abnormal behaviour identification under crowd scene, current research work is focused primarily upon using crowd movement's feature, these
Feature includes motion vector intensity and motion vector direction.Using following formula, motion vector intensity is corresponded to particle energy's
Each direction is modified:
Wherein riFor foreground area and the speed of background area, i is the quantity of video frame image, and j is the quantity of moving mass,
K is the quantity of motion vector.
The threshold value in each direction of motion vector intensityDepend onMaximum:
Step B2, the abnormal behaviour model learning based on dispersion and establishment.
The threshold value in each point direction of motion vector intensity is depended onMaximum:
The probability distribution of motion vector is represented by the probability that each motion vector occurs in the picture.One width is almost by complete
The image of the single intensity composition in portion, the information that it is included is less, i.e., its entropy is relatively low, now the probability distribution correspondence of motion vector
In the dispersion of the probability distribution of the single peak value of low entropy;Conversely, dispersion profile produces a higher entropy.The definition of entropy is such as
Under:
Wherein, X is probable value { x1,...,xnDiscrete random variable, N' is a probability distribution.Similarly, combination entropy
It is defined as follows:
Wherein, Y is probable value { y1,...,ynDiscrete random variable.
Dependence between mensurable two stochastic variable of interactive information, interactive information method is used to measure two direction histograms
Between similitude, the two direction histograms come from three adjacent image frames.Interactive information is the definition based on entropy:
I (X, Y)=H (X)+H (Y)-H (X, Y)
Wherein H (X) and H (Y) is X and the Shannon entropy of Y-direction respectively.
Above formula contains H (X, Y) item, it means that maximum interactive information is associated with the minimum of combination entropy.High
Interactive information represents uncertain a large amount of reductions;Low interactive information represents a less reduction;The zero of two stochastic variables
Interactive information is meant to be separate variable.The interactive information of revised particle energy and direction histogram contributes to different
Chang Hangwei detection.
In the present invention, direction histogram is set up by motion vector, for represent tracking target the direction of motion distribution and
Direction of motion trend.In addition, joint direction histogram can be set up by vector field as two continuous phases.The two vector fields by
Three consecutive frames are derived, wherein the Direction Probability distribution per frame is obtained by the segmentation of direction histogram vector summation entropy.Institute
The entropy for having frame of video is obtained by normalized interactive information.
Normalized mutual information represents the interactive information of normalized combination entropy, and it is defined as follows:
Wherein NMI has fixed lower bound 0, the fixed upper bound 1.When two cluster classifications are identical, it is worth for 1;When two clusters
When classification is separate, it is worth for 0.
Therefore, institute's extracting method of the present invention not only can extract two groups of characteristic vectors for calculated direction probability distribution, and can
Calculate the entropy based on the two probability distribution and the interactive information of combination entropy.
Step B3, based on the dynamic (dynamical) abnormal behaviour model learning of Lagrangian particle and establishment.
Handed over due to there is various systematic survey deviations, such as entropy and standard deviation, therefore the present invention using normalization
Mutual information and standard variance carry out the confusion degree of detection motion, because other methods are compared to, normalization interaction letter
Breath and standard variance can obtain more preferable overall performance.
Under the crowd is dense crowded environment, their motion conditions can be observed according to the space-time characteristic of crowd, are realized different
Reason condition is detected.Target following such as is carried out using the dynamic (dynamical) dynamical system concept of Lagrangian particle, the viewpoint comes from stream
The computational methods of body dynamics.
Crowd dynamics whether there is the suddenly change of motion state mainly for detection of a video sequence, and this is due to
When ignoring the direction of motion vector and only considering the intensity of motion vector, the effect of the unusual checking based on energy variance is simultaneously
It is not very good.Therefore, the present invention proposes normalized direction histogram interactive information.Normalized direction histogram interaction letter
The computing formula of breath is as follows:
Wherein γ ∈ [0,1] are fusion parameters, and σ represents standard deviation.
For the unusual checking under crowd scene, dynamic abnormal detection is very important, because dynamic
State abnormal behaviour is frequently accompanied by obvious change, situations such as assembling, disperse and be chaotic of such as crowd.Experimental result shows, this hair
Bright institute's extracting method can effectively recognize abnormal behaviour.
Finally, the step C carries out the global abnormal behavioral value of moving target according to learning outcome, determines abnormal area
And effectively marked.
Step C1, pretreatment, the pretreatment include filtering and smoothing processing.
It will appear from largely blocking during a large amount of target aggregations, this will increase unusual checking difficulty.Therefore, the present invention is used
3 × 3 median filter is handled motion feature, and then moving region is put down using 3 × 3 mean filter
It is sliding.
Motion smoothing device is used to carry out smoothly the target moved suddenly.To realize this target, it need to make on motion
The hypothesis of essence.Specifically, it is assumed that normal activity picture frame and interframe it is slowly varying be it is smooth, and motion artifacts with
The change of movement velocity and change.Therefore, the present invention carries out motion smoothing by moving average filter.
To calculate the optical flow characteristic of continuous interframe, reliable Optic flow information is quickly obtained using signature tracking algorithm.This
Invention will combine picture material variation characteristic and calculate Optic flow information with many grid. policies.
Step C2, unusual checking.
It is carried out in three steps detection:By taking the scene video image of hall as an example, using above-mentioned steps extract moving target feature and
Model is set up, it is normal behaviour such as Fig. 2 (a) shown that can obtain the moving object in video sequences under the scene of hall, and in hall
Moving object in video sequences is shown in abnormal behaviour such as Fig. 2 (b) under scene.
Then, after above-mentioned steps extract moving target feature and set up model, according to establishment based on energy, based on color
Dissipate and carried out based on the dynamic (dynamical) abnormal behaviour model of Lagrangian particle the global abnormal behavioral value of moving target.Work as foundation
After above-mentioned model, a behavioral value is carried out by taking scene under corridor as an example, is specifically included:First, the exception based on energy is utilized
Behavior model carries out unusual checking according to the motion vector strength characteristics of extraction;If in the absence of abnormal behaviour, utilizing base
Unusual checking carried out according to the histogrammic dispersion characteristic of the motion vector direction of extraction in the abnormal behaviour model of dispersion;If
In the absence of abnormal behaviour, then based on the dynamic (dynamical) abnormal behaviour model of Lagrangian particle according to the drawing between the space-time characteristic of extraction
Ge Lang particle dynamicses carry out unusual checking.When in the absence of abnormal area, the video image under the scene of corridor shows
Normal behaviour is shown as and without mark, shown in such as Fig. 3 (a);When there is abnormal area, determined according to the presence of abnormal behaviour different
Chang Hangwei regions, and be labeled with square, such as shown in Fig. 3 (b) and Fig. 3 (c).
Embodiments of the present invention are explained in detail above in conjunction with accompanying drawing, but the present invention is not limited to above-mentioned implementation
Mode, can also be on the premise of present inventive concept not be departed from the knowledge that those of ordinary skill in the art possess
Make a variety of changes.
Claims (6)
1. a kind of global abnormal behavioral value method based on target kinetic characteristic, it is characterised in that comprise the following steps:
Step A, the kinetic characteristic for extracting moving target in video;
Step B, extract according to step A kinetic characteristic moving target is carried out based on energy, based on dispersion and based on drawing respectively
The abnormal behaviour model learning of Ge Lang particle dynamicses and establishment;
Step C, entering based on energy, based on dispersion and based on the dynamic (dynamical) abnormal behaviour model of Lagrangian particle according to establishment
The global abnormal behavioral value of row moving target, determines abnormal area and is marked.
2. the global abnormal behavioral value method based on target kinetic characteristic according to claim 1, it is characterised in that described
The kinetic characteristic of moving target includes in step A:Space-time characteristic, motion vector characteristic, motion vector direction histogram, motion arrow
Measure intensity, motion vector mesh scale.
3. the global abnormal behavioral value method based on target kinetic characteristic according to claim 1, it is characterised in that described
Abnormal behaviour model learning based on energy in step B, be specially:
Determine the degree of contact of two particle potential energy in solid:
Wherein, U (r) represents the potential energy between two particles, and r represents their Euclidean distance, and a, b, m, n represents constant, and m>
n;
The interaction force between two particle is calculated in conjunction with power:
Kinetic characteristic is extracted with reference to step A, each direction of motion vector intensity is modified:
Wherein, riFor foreground area and the speed of background area, i is the quantity of video frame image, and j is the quantity of moving mass, and k is
The quantity of motion vector;W (r) represents connection weight of the distance for r two particles,For j-th of fortune in i-th of frame of video
The direction of k-th of motion vector intensity of motion block,For improved motion vector intensity;
The threshold value in each direction of the motion vector intensityDepend onMaximum:
Wherein, wiFor main Directional Window mouthful.
4. the global abnormal behavioral value method based on target kinetic characteristic according to claim 1, it is characterised in that:It is described
Abnormal behaviour model learning based on dispersion and establishment in step B, be specially:
Determine that threshold value of the motion vector intensity in each point of direction is depended onMaximum:
According to the definition of entropy and combination entropy, it is the definition based on entropy to determine interactive information:
I (X, Y)=H (X)+H (Y)-H (X, Y)
Wherein, H (X) and H (Y) is X and the Shannon entropy of Y-direction respectively;H (X, Y) is combination entropy;
The interactive information of normalized combination entropy is represented using normalized mutual information, it is defined as follows:
Wherein, NMI (X, Y) has fixed lower bound 0, the fixed upper bound 1.
5. the global abnormal behavioral value method based on target kinetic characteristic according to claim 1, it is characterised in that:It is described
The dynamic (dynamical) abnormal behaviour model learning of Lagrangian particle, including normalized direction histogram interaction letter are based in step B
The calculating of breath:
Wherein, γ ∈ [0,1] are fusion parameters, and σ represents standard deviation;NMI (X, Y) is the interactive information of combination entropy;For fortune
Threshold value of the dynamic vector intensity in each direction.
6. the global abnormal behavioral value method based on target kinetic characteristic according to claim 1, it is characterised in that:It is described
The global abnormal behavioral value of moving target is carried out in step C to be included:Moving target is pre-processed, the pretreatment includes
Filtering and smoothing processing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710092206.5A CN106951827A (en) | 2017-02-21 | 2017-02-21 | A kind of global abnormal behavioral value method based on target kinetic characteristic |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710092206.5A CN106951827A (en) | 2017-02-21 | 2017-02-21 | A kind of global abnormal behavioral value method based on target kinetic characteristic |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106951827A true CN106951827A (en) | 2017-07-14 |
Family
ID=59466496
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710092206.5A Pending CN106951827A (en) | 2017-02-21 | 2017-02-21 | A kind of global abnormal behavioral value method based on target kinetic characteristic |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106951827A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108764209A (en) * | 2018-06-11 | 2018-11-06 | 南京邮电大学 | A kind of personage's abnormal behaviour automatic testing method based on monitor video |
CN108846500A (en) * | 2018-04-17 | 2018-11-20 | 安徽师范大学 | Travel history data capture method based on Flickr geographical labels member |
CN113033504A (en) * | 2021-05-19 | 2021-06-25 | 广东众聚人工智能科技有限公司 | Multi-scale video anomaly detection method |
CN113554131A (en) * | 2021-09-22 | 2021-10-26 | 四川大学华西医院 | Medical image processing and analyzing method, computer device, system and storage medium |
-
2017
- 2017-02-21 CN CN201710092206.5A patent/CN106951827A/en active Pending
Non-Patent Citations (3)
Title |
---|
TIAN CAO 等: "Abnormal Crowd Motion Analysis", 《2009 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND BIOMIMETICS(ROBIO)》 * |
YUAN YUAN 等: "Online Anomaly Detection in Crowd Scenes via Structure Analysis", 《IEEE TRANSACTIONS ON CYBERNETICS》 * |
YUNYOUNG NAM 等: "Real-time abnormal situation detection based on particle advection in crowded scenes", 《JOURNAL OF REAL-TIME IMAGE PROCESSING》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108846500A (en) * | 2018-04-17 | 2018-11-20 | 安徽师范大学 | Travel history data capture method based on Flickr geographical labels member |
CN108764209A (en) * | 2018-06-11 | 2018-11-06 | 南京邮电大学 | A kind of personage's abnormal behaviour automatic testing method based on monitor video |
CN113033504A (en) * | 2021-05-19 | 2021-06-25 | 广东众聚人工智能科技有限公司 | Multi-scale video anomaly detection method |
CN113033504B (en) * | 2021-05-19 | 2021-08-27 | 广东众聚人工智能科技有限公司 | Multi-scale video anomaly detection method |
CN113554131A (en) * | 2021-09-22 | 2021-10-26 | 四川大学华西医院 | Medical image processing and analyzing method, computer device, system and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103745230B (en) | Adaptive abnormal crowd behavior analysis method | |
Wu et al. | A detection system for human abnormal behavior | |
EP1868162A2 (en) | Unusual action detection based on image analysis involving CHLAC feature extraction | |
Nallaivarothayan et al. | An MRF based abnormal event detection approach using motion and appearance features | |
CN106951827A (en) | A kind of global abnormal behavioral value method based on target kinetic characteristic | |
CN103955671B (en) | Human behavior recognition method based on rapid discriminant common vector algorithm | |
Ma et al. | Advanced local binary pattern descriptors for crowd estimation | |
CN105096342A (en) | Intrusion detection algorithm based on Fourier descriptor and histogram of oriented gradient | |
Gonzalez et al. | Context-independent facial action unit recognition using shape and gabor phase information | |
CN105809206A (en) | Pedestrian tracking method | |
Wang et al. | Online detection of abnormal events in video streams | |
CN108062557A (en) | Dimension self-adaption method for tracking target based on Fast Compression track algorithm | |
He et al. | Online unsupervised learning classification of pedestrian and vehicle for video surveillance | |
Li et al. | Car tracking algorithm based on Kalman filter and compressive tracking | |
Xia et al. | Anomaly detection in traffic surveillance with sparse topic model | |
Yang et al. | Crowd density and counting estimation based on image textural feature | |
Duan et al. | Crowded abnormal detection based on mixture of kernel dynamic texture | |
Xiao et al. | An Efficient Crossing-Line Crowd Counting Algorithm with Two-Stage Detection. | |
Hunter et al. | Exploiting sparse representations in very high-dimensional feature spaces obtained from patch-based processing | |
Zheng et al. | Traffic Sign Recognition Based on Learning Vector Quantization and Convolution Neural Network | |
Soeleman et al. | Tracking Moving Objects based on Background Subtraction using Kalman Filter | |
da Rocha et al. | A proposal for human action classification based on motion analysis and artificial neural networks | |
Revathi et al. | Hybridisation of feed forward neural network and self-adaptive PSO with diverse of features for anomaly detection | |
Bhaskar et al. | Background modeling using adaptive cluster density estimation for automatic human detection | |
Song et al. | Subsampling-based HOG for Multi-scale real-time Pedestrian Detection |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170714 |