CN108596028A - A kind of unusual checking algorithm based in video record - Google Patents
A kind of unusual checking algorithm based in video record Download PDFInfo
- Publication number
- CN108596028A CN108596028A CN201810224910.6A CN201810224910A CN108596028A CN 108596028 A CN108596028 A CN 108596028A CN 201810224910 A CN201810224910 A CN 201810224910A CN 108596028 A CN108596028 A CN 108596028A
- Authority
- CN
- China
- Prior art keywords
- point
- acceleration
- video
- crowd
- particle
- 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.)
- Granted
Links
- 230000001133 acceleration Effects 0.000 claims abstract description 69
- 238000000605 extraction Methods 0.000 claims abstract description 29
- 239000011159 matrix material Substances 0.000 claims abstract description 25
- 238000000034 method Methods 0.000 claims abstract description 24
- 206010000117 Abnormal behaviour Diseases 0.000 claims abstract description 15
- 238000010801 machine learning Methods 0.000 claims abstract description 12
- 230000003287 optical effect Effects 0.000 claims abstract description 6
- 230000001815 facial effect Effects 0.000 claims abstract description 5
- 239000002245 particle Substances 0.000 claims description 41
- 230000005856 abnormality Effects 0.000 claims description 17
- 230000014509 gene expression Effects 0.000 claims description 16
- 238000012544 monitoring process Methods 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 9
- 230000002547 anomalous effect Effects 0.000 claims description 8
- 238000009826 distribution Methods 0.000 claims description 8
- 230000008921 facial expression Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000006073 displacement reaction Methods 0.000 claims description 6
- 238000013459 approach Methods 0.000 claims description 4
- 238000001514 detection method Methods 0.000 abstract description 9
- 230000002159 abnormal effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000004075 alteration Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012549 training Methods 0.000 description 1
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
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/254—Analysis of motion involving subtraction of images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30196—Human being; Person
-
- 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
- G06V20/44—Event detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of unusual checking algorithms based in video record, belong to intelligent video-detect early warning field.Video is carried out foreground extraction first and carries out mesh generation to crowd by the present invention, using optical flow method calculating speed matrix and then obtains acceleration matrix, the center monitors point of crowd's exception is determined further according to the size and Orientation of acceleration;Then the facial feature extraction of face in machine learning carry out crowd is recycled;The center monitors point finally being had identified that according to the threshold contrast that machine learning is trained is detected, and then judges whether the crowd of the Spot detection point is in abnormal behaviour.It is hit cruelly if detecting to have occurred in crowd, assembles a crowd to make trouble and when the abnormal behaviours such as similar dangerous event, video surveillance will alarm to this.The present invention not only overcomes due to abnormal behaviour situation complicated difficult the problem of to be detected, but also can predict the generation of abnormal behaviour to a certain extent.
Description
Technical field
The present invention relates to a kind of unusual checking algorithms based in video record, belong to intelligent video-detect early warning skill
Art field.
Background technology
With the continuous development of society, the place of many crowd massings easy tos produce safety problem, for example violence is asked safely
Topic, the various events such as group affray and robbery are all existing security risks.Therefore video monitoring just becomes and ensures public field
The essential equipment of institute's safety, the presence of video monitoring system reduce the probability of time generation to a certain extent, still
It equally spends human and material resources, staff is needed sporadically to watch video and then could find locale.
Invention content
The technical problem to be solved in the present invention is to provide a kind of unusual checking algorithms based in video record, provide
It is a kind of rationally, the screening scheme that real-time is good, accuracy is high, not only overcome that monitoring range is wide, and abnormal behaviour pattern is complicated
The problem of being difficult to be detected, but also the generation of abnormal behaviour can be predicted to a certain extent, alarm can be played
Effect, improves the scope of application and the effect of monitoring.
The technical solution adopted by the present invention is:A kind of unusual checking algorithm based in video record, including it is as follows
Step:
Step1 carries out foreground extraction for the video record provided;
Step2 carries out gridding processing for Step1 treated video records;
Step3 is carried out feature point extraction to Step2 treated video records, and is tracked using optical flow method, is obtained
To rate matrices;
Step4 obtains the size a and deflection β of acceleration, and then obtains acceleration matrix;
Step5 obtains the same particle different moments acceleration in adjacent two field pictures by analyzing the deflection of acceleration
Direction change angulation β ' is spent, and particle point proportion m in each region is obtained by the region of β ' distributions;It obtains
Acceleration a and accounting m is compared with the a* and m* set before respectively, if | a | > | a*| and m < m*Can by this point
Position judgment be abnormal behaviour spot O1Point, as center monitors point, O1Point coordinates is (xo1,yo1), a* is to be arranged in advance
Crowd it is normal when peak acceleration, m* is the reduced parameter being arranged in advance, m < m*When illustrate crowd's neither one determine
Acceleration direction is in rambling state, and abnormal behaviour is not present;
Step6 establishes model by machine learning, and model is to be trained to obtain according to classical case and a large amount of data
Crowd's changing ratio standard value ρ*;
Step7, it is ρ to define expression shape change rate first
Indicate that changing features degree is more than the number of the general value upper limit in n expression crowds, N indicates that the target group of extraction is total
Number,
Then facial characteristics is carried out to the processed video record of Step2 treated griddings according to classical approach to carry
It takes, expression shape change rate ρ will be obtained in the model of the facial expression input Step6 foundation of extraction, ρ and ρ will be gone out*Value compare,
If ρ is more than ρ*Then think this region point O of extraction human face's expression in the video2In exception, O2Point coordinates is (xo2,
yo2);
Step8, by O2Point coordinates (xo2,yo2) and O1Point coordinates (xo1,yo1) compared, allow model in certain error
It is considered that 2 points of expressions are the same spots in enclosing, then monitoring point i.e. anomalous event centered on the point can be specified
Point O occurs.
Specifically, the step Step1 is specifically included:
Foreground is extracted using frame difference method, target object and background are distinguished;
Frame difference method is operated using the image sequence between consecutive frame, is taken absolutely to the gray value of the pixel of image
Value, then by being compared to have obtained sport foreground information with threshold value, also it is achieved that the extraction of foreground, the figure of consecutive frame
Aberration score value calculates as follows
D (x, y, i)=| I (x, y, i+1)-I (x, y, i) | (1-1)
Wherein D (x, y, i) indicates that the difference value of image, I (x, y, i+1) indicate the gray value of i+1 frame, I (x, y, i) table
The gray value for showing the i-th frame, the pixel value of certain point is more than some preset threshold value after image difference, illustrates that the point is at this time
Belong to foreground point, otherwise belongs to background dot.
Specifically, the step of video record gridding is handled in the step Step2 specifically includes:
Step2.1:Video is extracted in the frame picture of t moment image, is denoted as Ft;
Step2.2:By entire frame picture according to p1*p2To be divided into a series of sub-box.
Specifically, the step Step3 is specifically included:
Step3.1:Calculating speed size vt
A certain moving foreground object is chosen on the basis of Step2 ready-portioned small grids as feature extraction target, vacation
If target A is a particle, in t moment particle position pixel coordinate AtFor (xt,yt), passing through time τ, corresponding t+ τ
The frame picture F at momentt+τParticle position pixel coordinate At+τFor (xt+τ,yt+τ);
Step3.1.1:By the coordinate relationship of particle position between adjacent two frame it is found that particle point AtIt is horizontal and vertical
Displacement on direction is respectively (1-2), (1-3):
Δ x=xt+τ-xt (1-2)
Δ y=yt+τ-yt (1-3)
It can be obtained in t moment A according to vector calculation methodtPoint velocity magnitude be
Utilize (1-4) that the velocity magnitude v at the point moment can be calculatedt;
Step3.1.2:Approximating assumption τ is 0, be can be ignored because the time interval of consecutive frame is very short, in this way can be with
(1-4) formula is reduced to the form of (1-5)
Step3.2:Calculating speed direction θt
Using in Step3.1.1 coordinate and displacement (1-2) and (1-3) the direction θ of speed is calculatedt, by speed
Vector and the expression of the angle of horizontal direction,
θt=arctan (Δ y/ Δs x) (1-6)
θtValue range is [- π, π]
Step3.3:One is carried out to each characteristic point in video pictures frame according to the step of Step3.1 and Step3.2
One processing, obtains corresponding velocity magnitude matrix and directional velocity matrix.
Specifically, the step of acquiring acceleration matrix in the Step4 is as follows:
Step4.1:Calculate the size a of acceleration
It can be calculated in t moment A according to (1-4) formula in Step3tThe velocity magnitude of point is vt, after time τ again
Calculate in t+ τ moment A according to (1-4) formulat+τThe velocity magnitude of point is vt+τ, the variable quantity of speed is in time τ
Δ v=vt+τ-vt (1-7)
Can obtain acceleration magnitude according to the calculation formula of acceleration is
A=Δ v/ τ (1-8)
Again τ can approximation ignore, it is possible to be approximately considered acceleration a=Δs v
It is obtained by quadrilateral rule
And
Step4.2:Calculate the direction β of acceleration
β=arctan (Δ vy/Δvx) (1-11)
β value ranges are [- π, π]
Step4.3:One is carried out to each particle point in video pictures frame according to the step of Step4.1 and Step4.2
One processing, obtains corresponding acceleration magnitude matrix and acceleration direction matrix.
Specifically, the Step5 specifically comprises the following steps:
Step5.1:Wherein a* is the peak acceleration when crowd that is arranged in advance is normal, shows crowd when a is more than a*
Acceleration be in abnormality, judge again at this time acceleration direction whether be in abnormality;
Step5.2:β ' is the same particle different moments acceleration direction change angulation in adjacent two field pictures,
M is to pass through the region that β ' is distributed to obtain particle point proportion in each region;2 π are divided into four quadrants, each quadrant is again
Two regions are divided into, according to the acceleration direction matrix of present frame particle point, count the number of particle point in each minizone
Amount accounts for the ratio m of total numberi, work as miMore than preset m*When illustrate that the flow direction of majority of populations is consistent, people
Group can be regarded as in normal condition, otherwise can regard as in abnormality;
Step5.3:When acceleration magnitude and direction are in abnormality it may determine that location point O1(xo1,yo1)
Centered on monitoring point.
Specifically, the Step8 is comprised the concrete steps that:
The O that Step5 and Step7 are obtained1(xo1,yo1) and O2(xo2,yo2) two point coordinates are analyzed,
Work as O1And O2It can think that the two center monitors points are the same position when distance d≤d* of two central points,
D* indicates the upper threshold set in advance, finally according to O1And O2Determine that central point O, O point coordinates is (x0,yo)
It is determined that point O occurs for anomalous event from above.
The beneficial effects of the invention are as follows:
1, patent of the present invention, realizes a kind of unusual checking algorithm based in video record, which combines
Optical flow method calculates acceleration to judge the abnormal target for realizing crowd's abnormality detection with the detection again of machine learning of crowd, carries
High accuracy.
2, patent of the present invention is largely capable of the prompting staff progress early warning processing of intelligence, reduces society
The generation of security incident.
The beneficial effects of the invention are as follows:
(1) method of the invention can estimate cycle length;
(2) method of the invention is capable of detecting when period and paracycle and three kinds of states of chaos;
(3) method of the invention can change parameter and the movement convenient for observing time sequence at any time in calculating process
State and process.
Description of the drawings
Fig. 1 is the particular flow sheet in the present invention;
Fig. 2 is that crowd flows simple distribution map in Step5 of the present invention.
Fig. 3 is machine learning unusual checking modeling process figure of the present invention;
Fig. 4 is the flow chart that study model part is established in machine learning unusual checking model of the present invention.
Specific implementation mode
In the following with reference to the drawings and specific embodiments, the invention will be further described.
Embodiment 1:As shown in Figs 1-4, a kind of unusual checking algorithm based in video record, including walk as follows
Suddenly:
Step1 carries out foreground extraction for the video record provided;
Step2 carries out gridding processing for Step1 treated video records;
Step3 is carried out feature point extraction to Step2 treated video records, and is tracked using optical flow method, is obtained
To rate matrices;
Step4 obtains the size a and deflection β of acceleration, and then obtains acceleration matrix;
Step5 obtains the same particle different moments acceleration in adjacent two field pictures by analyzing the deflection of acceleration
Direction change angulation β ' is spent, and the region for passing through β ' distributions obtains particle point proportion m in each region;It obtains
Acceleration a and accounting m is compared with the a* and m* set before respectively, if | a | > | a*| and m < m*Can by this point
Position judgment be abnormal behaviour spot O1Point, as center monitors point, O1Point coordinates is (xo1,yo1), a* is to be arranged in advance
Crowd it is normal when peak acceleration, m* is the reduced parameter being arranged in advance, m < m*When illustrate crowd's neither one determine
Acceleration direction is in rambling state, and abnormal behaviour is not present;
M* is a reduced parameter, has been set in advance, generally according to being set as 50% for routine, is made with it
Comparison be front and back two frame speeds direction change β ' size proportion m in the zone,
Step6 establishes model by machine learning, and model is to be trained to obtain according to classical case and a large amount of data
Crowd changing ratio standard value ρ *;
Step7, it is ρ to define expression shape change rate first
Indicate that changing features degree is more than the number of the general value upper limit in n expression crowds, N indicates that the target group of extraction is total
Number,
Then facial characteristics is carried out to the processed video record of Step2 treated griddings according to classical approach to carry
It takes, expression shape change rate ρ will be obtained in the model of the facial expression input Step6 foundation of extraction, the value for going out ρ and ρ * is compared,
This region point O of extraction human face's expression in the video is thought if ρ is more than ρ *2In exception, O2Point coordinates is (xo2,
yo2);
Step8, by O2Point coordinates (xo2,yo2) and O1Point coordinates (xo1,yo1) compared, allow model in certain error
It is considered that 2 points of expressions are the same spots in enclosing, then monitoring point i.e. anomalous event centered on the point can be specified
Point O occurs.
Further, the step Step1 is specifically included:
Foreground is extracted using frame difference method, target object and background are distinguished;
Frame difference method is operated using the image sequence between consecutive frame, is taken absolutely to the gray value of the pixel of image
Value, then by being compared to have obtained sport foreground information with threshold value, also it is achieved that the extraction of foreground, the figure of consecutive frame
Aberration score value calculates as follows
D (x, y, i)=| I (x, y, i+1)-I (x, y, i) | (1-1)
Wherein D (x, y, i) indicates that the difference value of image, I (x, y, i+1) indicate the gray value of i+1 frame, I (x, y, i) table
The gray value for showing the i-th frame, the pixel value of certain point is more than some preset threshold value after image difference, illustrates that the point is at this time
Belong to foreground point, otherwise belongs to background dot.
Foreground is extracted using frame difference method, light sensitive degree of the frame difference method used in the present invention for environment
It is smaller, so influence of the environment to result is reduced to a certain extent, it can be by the wheel of foreground target by frame difference method
Exterior feature extracts.Foreground extraction is the basis of carry out crowd's abnormality detection, and target object and background are only distinguished ability
Preferably carry out the mesh generation of next step video record.Calculus of finite differences be used for video frame between arithmetic speed faster.
Further, the step of video record gridding is handled in the step Step2 specifically includes:
Step2.1:Video is extracted in the frame picture of t moment image, is denoted as Ft;
Step2.2:By entire frame picture according to p1*p2To be divided into a series of sub-box.
Specifically, the step Step3 is specifically included:
Step3.1:Calculating speed size vt
A certain moving foreground object is chosen on the basis of Step2 ready-portioned small grids as feature extraction target, vacation
If target A is a particle, in t moment particle position pixel coordinate AtFor (xt,yt), passing through time τ, corresponding t+ τ
The frame picture F at momentt+τParticle position pixel coordinate At+τFor (xt+τ,yt+τ);
Step3.1.1:By the coordinate relationship of particle position between adjacent two frame it is found that particle point AtIt is horizontal and vertical
Displacement on direction is respectively (1-2), (1-3):
Δ x=xt+τ-xt (1-2)
Δ y=yt+τ-yt (1-3)
It can be obtained in t moment A according to vector calculation methodtPoint velocity magnitude be
Utilize (1-4) that the velocity magnitude v at the point moment can be calculatedt;
Step3.1.2:Approximating assumption τ is 0, be can be ignored because the time interval of consecutive frame is very short, in this way can be with
(1-4) formula is reduced to the form of (1-5)
Step3.2:Calculating speed direction θt
Using in Step3.1.1 coordinate and displacement (1-2) and (1-3) the direction θ of speed is calculatedt, by speed
Vector and the expression of the angle of horizontal direction,
θt=arctan (Δ y/ Δs x) (1-6)
θtValue range is [- π, π]
Step3.3:One is carried out to each characteristic point in video pictures frame according to the step of Step3.1 and Step3.2
One processing, obtains corresponding velocity magnitude matrix and directional velocity matrix.
Further, the step of acquiring acceleration matrix in the Step4 is as follows:
Step4.1:Calculate the size a of acceleration
It can be calculated in t moment A according to (1-4) formula in Step3tThe velocity magnitude of point is vt, after time τ again
Calculate in t+ τ moment A according to (1-4) formulat+τThe velocity magnitude of point is vt+τ, the variable quantity of speed is in time τ
Δ v=vt+τ-vt (1-7)
Can obtain acceleration magnitude according to the calculation formula of acceleration is
A=Δ v/ τ (1-8)
Again τ can approximation ignore, it is possible to be approximately considered acceleration a=Δs v
It is obtained by quadrilateral rule
And
Step4.2:Calculate the direction β of acceleration
β=arctan (Δ vy/Δvx) (1-11)
β value ranges are [- π, π]
Step4.3:One is carried out to each particle point in video pictures frame according to the step of Step4.1 and Step4.2
One processing, obtains corresponding acceleration magnitude matrix and acceleration direction matrix.
Further, the Step5 specifically comprises the following steps:
Step5.1:Wherein a* is the peak acceleration when crowd that is arranged in advance is normal, shows crowd when a is more than a*
Acceleration be in abnormality, judge again at this time acceleration direction whether be in abnormality;
Step5.2:β ' is the same particle different moments acceleration direction change angulation in adjacent two field pictures,
M is to pass through the region that β ' is distributed to obtain particle point proportion in each region;2 π are divided into four quadrants, each quadrant is again
Two regions are divided into, according to the acceleration direction matrix of present frame particle point, count the number of particle point in each minizone
Amount accounts for the ratio m of total numberi, work as miMore than preset m*When illustrate that the flow direction of majority of populations is consistent, people
Group can be regarded as in normal condition, otherwise can regard as in abnormality;
Step5.3:When acceleration magnitude and direction are in abnormality it may determine that location point O1(xo1,yo1)
Centered on monitoring point.
Further, the Step8 is comprised the concrete steps that:
The O that Step5 and Step7 are obtained1(xo1,yo1) and O2(xo2,yo2) two point coordinates are analyzed,
Work as O1And O2It can think that the two center monitors points are the same position when distance d≤d* of two central points,
D* indicates the upper threshold set in advance, finally according to O1And O2Determine that central point O, O point coordinates is (x0,yo)
It is determined that point O occurs for anomalous event from above.
In Step6, the accuracy present invention in order to improve time prediction proposes the method based on machine learning, passes through machine
Model is established in device study, and the foundation of model is according to classical case and as obtained by a large amount of data analysis and experiment;Video
Include that there is a phenomenon where under normal circumstances, these videos are first specifically divided into two, a part is used for for abnormal behaviour
The foundation of detection model, a part are used for the correctness of detection model, are optimized further according to the result of detection, to reach model
Accuracy.
In Step7, carry out judging that point occurs for anomalous event again using the model being previously mentioned in Step6, first according to classical approach
Facial feature extraction is carried out to the processed video record of gridding in video, the facial expression of people is in face of different things
When variation be it is bigger, when occur such as dangerous sudden and violent of accident hit event and assemble a crowd to make trouble event when micro- table
End of love or prodigious.Further judge O by expression shape change ratio ρ in the extraction of facial expression and crowd2Point
Occur with the presence or absence of anomalous event;
As shown in Figure 1, carrying out foreground extraction after getting frame video pictures first, processing that then video network will be formatted will
Target regards particle point as, carries out feature extraction to the speed of particle point and direction using optical flow method, and then accelerated accordingly
Matrix is spent, judges monitoring point O by the size and Orientation variation for analyzing acceleration and corresponding distributed layout1Whether
For target point, that is, Spot detection point.In order to improve accuracy, simple model is established using machine learning, by the face for detecting face
The micro- expression shape change accounting rate ρ in portion and numerical value relatively obtain monitoring point O2Whether it is abnormal point, if two monitoring points are
Abnormal point, according to the position of the further center monitors point of 2 points of coordinate positions.
Fig. 2 indicates the acceleration of the direction of motion and acquiring size by particle, when judging the distribution of its direction change
Simple distribution map:Entire plane space is divided into four quadrants, then four quadrants are respectively divided into two regions, according to phase
The particle point that the adjacent same particle point acceleration direction change angle of two frames is fallen into the distribution and each quadrant in space
Number accounts for the ratio m of sum to judge the acceleration of crowd whether in abnormal.Illustrate major part when m is more than preset m*
The flow direction of crowd is consistent, orderly movement, be can be regarded as in normal condition, otherwise can be regarded place as
In abnormality.
Fig. 3 indicates machine learning unusual checking modeling process, and the classical case video of importing is divided
Class is divided into two large divisions, and a part is used for the foundation of model, and a part is for detecting whether the model having had built up can be sentenced
It is disconnected correct;Undoubtedly will appear during detection error or other the problem of, learning model is optimized accordingly, with
Reach the set goal.
Fig. 4 indicates that the process that study model part is established in machine learning unusual checking model, video are starting just
Whether it is in abnormality on identified, known video is imported to the rudimentary model made according to face feature, root
According to the comparison and analysis of the target signature in video, preliminary assessment is carried out to picture, judges whether to be in abnormality, warp
Continuous test and training are crossed, the parameter in model is optimized and changed, learning model is finally obtained.
The present invention provides a kind of screening schemes reasonable, real-time is good, accuracy is high, not only overcome monitoring range
Extensively, the problem of abnormal behaviour pattern complicated difficult is to be detected, but also abnormal behaviour can be predicted to a certain extent
Occur, improves the scope of application and the effect of intelligent monitoring.The present invention monitors the exception in video using intelligent electronic equipment
Then behavior alarms to relevant staff, thus can effectively be avoided using intelligent video monitoring system sudden and violent
Lixing be etc. a series of security risks generation, save manpower, material resources and financial resources.
The specific implementation mode of the present invention is explained in detail above in conjunction with attached drawing, but the present invention is not limited to above-mentioned
Embodiment within the knowledge of a person skilled in the art can also be before not departing from present inventive concept
Put that various changes can be made.
Claims (7)
1. a kind of unusual checking algorithm based in video record, it is characterised in that:Include the following steps:
Step1 carries out foreground extraction for the video record provided;
Step2 carries out gridding processing for Step1 treated video records;
Step3 is carried out feature point extraction to Step2 treated video records, and is tracked using optical flow method, and speed is obtained
Spend matrix;
Step4 obtains the size a and deflection β of acceleration, and then obtains acceleration matrix;
Step5 obtains acceleration side of the same particle different moments in adjacent two field pictures by analyzing the deflection of acceleration
To variation angulation β ', and the region for passing through β ' distributions obtains particle point proportion m in each region;Obtained acceleration
Degree a and accounting m is compared with the a* and m* set before respectively, if | a | > | a*| and m < m*It can be by the position
It is judged as YES abnormal behaviour spot O1Point, as center monitors point, O1Point coordinates is (xo1,yo1), a* is the people being arranged in advance
Peak acceleration when group is normal, m* is the reduced parameter being arranged in advance, m<m*When illustrate crowd's neither one determine acceleration
Direction is spent, rambling state is in, abnormal behaviour is not present;
Step6 establishes model by machine learning, and model is to be trained to obtain crowd according to classical case and a large amount of data
Changing ratio standard value ρ*;
Step7, it is ρ to define expression shape change rate first
Indicate that changing features degree is more than the number of the general value upper limit in n expression crowds, N indicates the total people of the target group of extraction
Number,
Then facial feature extraction is carried out to the processed video record of Step2 treated griddings according to classical approach,
Expression shape change rate ρ will be obtained in the model of the facial expression input Step6 foundation of extraction, ρ and ρ will be gone out*Value compare, if ρ
More than ρ*Then think this region point O of extraction human face's expression in the video2In exception, O2Point coordinates is (xo2,yo2);
Step8, by O2Point coordinates (xo2,yo2) and O1Point coordinates (xo1,yo1) compared, it can in certain allowable range of error
To think at 2 points, indicate is the same spot, then can specify monitoring point i.e. anomalous event centered on the point and point occurs
O。
2. a kind of unusual checking algorithm based in video record according to claim 1, it is characterised in that:It is described
Step Step1 specifically include:
Foreground is extracted using frame difference method, target object and background are distinguished;
Frame difference method is operated using the image sequence between consecutive frame, is taken absolute value to the gray value of the pixel of image,
Again by being compared to have obtained sport foreground information with threshold value, it is also achieved that the extraction of foreground, the image of consecutive frame
Difference value calculates as follows
D (x, y, i)=| I (x, y, i+1)-I (x, y, i) | (1-1)
Wherein D (x, y, i) indicates that the difference value of image, I (x, y, i+1) indicate that the gray value of i+1 frame, I (x, y, i) indicate the
The gray value of i frames, the pixel value of certain point is more than some preset threshold value after image difference, illustrates that the point is to belong at this time
Otherwise foreground point belongs to background dot.
3. a kind of unusual checking algorithm based in video record according to claim 1, it is characterised in that:It is described
Step Step2 in video record gridding handle the step of specifically include:
Step2.1:Video is extracted in the frame picture of t moment image, is denoted as Ft;
Step2.2:By entire frame picture according to p1*p2To be divided into a series of sub-box.
4. a kind of unusual checking algorithm based in video record according to claim 1, it is characterised in that:It is described
Step Step3 specifically include:
Step3.1:Calculating speed size vt
A certain moving foreground object is chosen as feature extraction target on the basis of Step2 ready-portioned small grids, it is assumed that should
Target A is a particle, in t moment particle position pixel coordinate AtFor (xt,yt), passing through time τ, corresponding t+ τ moment
Frame picture Ft+τParticle position pixel coordinate At+τFor (xt+τ,yt+τ);
Step3.1.1:By the coordinate relationship of particle position between adjacent two frame it is found that particle point AtBoth horizontally and vertically
On displacement be respectively (1-2), (1-3):
Δ x=xt+τ-xt (1-2)
Δ y=yt+τ-yt (1-3)
It can be obtained in t moment A according to vector calculation methodtPoint velocity magnitude be
Utilize (1-4) that the velocity magnitude v at the point moment can be calculatedt;
Step3.1.2:Approximating assumption τ is 0, be can be ignored because the time interval of consecutive frame is very short, in this way can be by (1-
4) formula is reduced to the form of (1-5)
Step3.2:Calculating speed direction θt
Using in Step3.1.1 coordinate and displacement (1-2) and (1-3) the direction θ of speed is calculatedt, by velocity vector with
The angle expression of horizontal direction,
θt=arctan (Δ y/ Δs x) (1-6)
θtValue range is [- π, π]
Step3.3:Each characteristic point in video pictures frame is located one by one according to the step of Step3.1 and Step3.2
Reason, obtains corresponding velocity magnitude matrix and directional velocity matrix.
5. a kind of unusual checking algorithm based in video record according to claim 1, it is characterised in that:It is described
Step4 in the step of acquiring acceleration matrix it is as follows:
Step4.1:Calculate the size a of acceleration
It can be calculated in t moment A according to (1-4) formula in Step3tThe velocity magnitude of point is vt, after time τ further according to
(1-4) formula calculates in t+ τ moment At+τThe velocity magnitude of point is vt+τ, the variable quantity of speed is in time τ
Δ v=vt+τ-vt (1-7)
Can obtain acceleration magnitude according to the calculation formula of acceleration is
A=Δ v/ τ (1-8)
Again τ can approximation ignore, it is possible to be approximately considered acceleration a=Δs v
It is obtained by quadrilateral rule
And
Step4.2:Calculate the direction β of acceleration
β=arctan (Δ vy/Δvx) (1-11)
β value ranges are [- π, π]
Step4.3:Each particle point in video pictures frame is located one by one according to the step of Step4.1 and Step4.2
Reason, obtains corresponding acceleration magnitude matrix and acceleration direction matrix.
6. a kind of unusual checking algorithm based in video record according to claim 1, it is characterised in that:It is described
Step5 specifically comprise the following steps:
Step5.1:Wherein a* is the peak acceleration when crowd that is arranged in advance is normal, show when a is more than a* crowd's plus
Speed has been in abnormality, judges whether the direction of acceleration is in abnormality again at this time;
Step5.2:β ' is that the same particle different moments acceleration direction change angulation, m are in adjacent two field pictures
The region for passing through β ' distributions obtains particle point proportion in each region;2 π are divided into four quadrants, each quadrant is drawn again
It is divided into two regions, according to the acceleration direction matrix of present frame particle point, counts the quantity of particle point in each minizone
Account for the ratio m of total numberi, work as miMore than preset m*When illustrate that the flow direction of majority of populations is consistent, crowd
It can be regarded as in normal condition, otherwise can regard as in abnormality;
Step5.3:When acceleration magnitude and direction are in abnormality it may determine that location point O1(xo1,yo1) be
Heart monitoring point.
7. a kind of unusual checking algorithm based in video record according to claim 1, it is characterised in that:It is described
Step8 comprise the concrete steps that:
The O that Step5 and Step7 are obtained1(xo1,yo1) and O2(xo2,yo2) two point coordinates are analyzed,
Work as O1And O2It can think that the two center monitors points are the same position when distance d≤d* of two central points,
D* indicates the upper threshold set in advance, finally according to O1And O2Determine that central point O, O point coordinates is (x0,yo)
It is determined that point O occurs for anomalous event from above.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810224910.6A CN108596028B (en) | 2018-03-19 | 2018-03-19 | Abnormal behavior detection algorithm based on video recording |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810224910.6A CN108596028B (en) | 2018-03-19 | 2018-03-19 | Abnormal behavior detection algorithm based on video recording |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108596028A true CN108596028A (en) | 2018-09-28 |
CN108596028B CN108596028B (en) | 2022-02-08 |
Family
ID=63626549
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810224910.6A Active CN108596028B (en) | 2018-03-19 | 2018-03-19 | Abnormal behavior detection algorithm based on video recording |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108596028B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110225299A (en) * | 2019-05-06 | 2019-09-10 | 平安科技(深圳)有限公司 | Video monitoring method, device, computer equipment and storage medium |
CN110274590A (en) * | 2019-07-08 | 2019-09-24 | 哈尔滨工业大学 | A kind of violent action detection method and system based on decision tree |
CN110298327A (en) * | 2019-07-03 | 2019-10-01 | 北京字节跳动网络技术有限公司 | A kind of visual effect processing method and processing device, storage medium and terminal |
CN111368089A (en) * | 2018-12-25 | 2020-07-03 | 中国移动通信集团浙江有限公司 | Service processing method and device based on knowledge graph |
CN111695404A (en) * | 2020-04-22 | 2020-09-22 | 北京迈格威科技有限公司 | Pedestrian falling detection method and device, electronic equipment and storage medium |
CN111814775A (en) * | 2020-09-10 | 2020-10-23 | 平安国际智慧城市科技股份有限公司 | Target object abnormal behavior identification method, device, terminal and storage medium |
WO2021134982A1 (en) * | 2019-12-31 | 2021-07-08 | 上海依图网络科技有限公司 | Video analysis-based event prediction method and device, and medium and system thereof |
CN114821808A (en) * | 2022-05-18 | 2022-07-29 | 湖北大学 | Attack behavior early warning method and system |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104732236A (en) * | 2015-03-23 | 2015-06-24 | 中国民航大学 | Intelligent crowd abnormal behavior detection method based on hierarchical processing |
CN105608440A (en) * | 2016-01-03 | 2016-05-25 | 复旦大学 | Minimum -error-based feature extraction method for face microexpression sequence |
KR20170051196A (en) * | 2015-10-29 | 2017-05-11 | 주식회사 세코닉스 | 3-channel monitoring apparatus for state of vehicle and method thereof |
CN107483887A (en) * | 2017-08-11 | 2017-12-15 | 中国地质大学(武汉) | The early-warning detection method of emergency case in a kind of smart city video monitoring |
-
2018
- 2018-03-19 CN CN201810224910.6A patent/CN108596028B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104732236A (en) * | 2015-03-23 | 2015-06-24 | 中国民航大学 | Intelligent crowd abnormal behavior detection method based on hierarchical processing |
KR20170051196A (en) * | 2015-10-29 | 2017-05-11 | 주식회사 세코닉스 | 3-channel monitoring apparatus for state of vehicle and method thereof |
CN105608440A (en) * | 2016-01-03 | 2016-05-25 | 复旦大学 | Minimum -error-based feature extraction method for face microexpression sequence |
CN107483887A (en) * | 2017-08-11 | 2017-12-15 | 中国地质大学(武汉) | The early-warning detection method of emergency case in a kind of smart city video monitoring |
Non-Patent Citations (3)
Title |
---|
PENG WANG ET AL.: "Automated video-based facial expression analysis of neuropsychiatric disorders", 《JOURNAL OF NEUROSCIENCE METHODS》 * |
刘皓: "基于条件随机场模型的异常行为检测方法研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
华斌等: "公共场所人群加速度异常检测系统", 《安全与环境学报》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111368089A (en) * | 2018-12-25 | 2020-07-03 | 中国移动通信集团浙江有限公司 | Service processing method and device based on knowledge graph |
CN111368089B (en) * | 2018-12-25 | 2023-04-25 | 中国移动通信集团浙江有限公司 | Business processing method and device based on knowledge graph |
CN110225299B (en) * | 2019-05-06 | 2022-03-04 | 平安科技(深圳)有限公司 | Video monitoring method and device, computer equipment and storage medium |
CN110225299A (en) * | 2019-05-06 | 2019-09-10 | 平安科技(深圳)有限公司 | Video monitoring method, device, computer equipment and storage medium |
CN110298327A (en) * | 2019-07-03 | 2019-10-01 | 北京字节跳动网络技术有限公司 | A kind of visual effect processing method and processing device, storage medium and terminal |
CN110298327B (en) * | 2019-07-03 | 2021-09-03 | 北京字节跳动网络技术有限公司 | Visual special effect processing method and device, storage medium and terminal |
CN110274590A (en) * | 2019-07-08 | 2019-09-24 | 哈尔滨工业大学 | A kind of violent action detection method and system based on decision tree |
WO2021134982A1 (en) * | 2019-12-31 | 2021-07-08 | 上海依图网络科技有限公司 | Video analysis-based event prediction method and device, and medium and system thereof |
CN111695404A (en) * | 2020-04-22 | 2020-09-22 | 北京迈格威科技有限公司 | Pedestrian falling detection method and device, electronic equipment and storage medium |
CN111695404B (en) * | 2020-04-22 | 2023-08-18 | 北京迈格威科技有限公司 | Pedestrian falling detection method and device, electronic equipment and storage medium |
CN111814775A (en) * | 2020-09-10 | 2020-10-23 | 平安国际智慧城市科技股份有限公司 | Target object abnormal behavior identification method, device, terminal and storage medium |
CN111814775B (en) * | 2020-09-10 | 2020-12-11 | 平安国际智慧城市科技股份有限公司 | Target object abnormal behavior identification method, device, terminal and storage medium |
CN114821808A (en) * | 2022-05-18 | 2022-07-29 | 湖北大学 | Attack behavior early warning method and system |
CN114821808B (en) * | 2022-05-18 | 2023-05-26 | 湖北大学 | Attack behavior early warning method and system |
Also Published As
Publication number | Publication date |
---|---|
CN108596028B (en) | 2022-02-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108596028A (en) | A kind of unusual checking algorithm based in video record | |
CN104123544B (en) | Anomaly detection method and system based on video analysis | |
CN107679471B (en) | Indoor personnel air post detection method based on video monitoring platform | |
CN105100757B (en) | Detector with integral structure | |
CN106128053A (en) | A kind of wisdom gold eyeball identification personnel stay hover alarm method and device | |
CN103456024B (en) | A kind of moving target gets over line determination methods | |
KR102144531B1 (en) | Method for automatic monitoring selectively based in metadata of object employing analysis of images of deep learning | |
CN102799893A (en) | Method for processing monitoring video in examination room | |
CN106127814A (en) | A kind of wisdom gold eyeball identification gathering of people is fought alarm method and device | |
CN110633643A (en) | Abnormal behavior detection method and system for smart community | |
CN103198296A (en) | Method and device of video abnormal behavior detection based on Bayes surprise degree calculation | |
CN103456009B (en) | Object detection method and device, supervisory system | |
CN102214359A (en) | Target tracking device and method based on hierarchic type feature matching | |
CN111488803A (en) | Airport target behavior understanding system integrating target detection and target tracking | |
CN106548142A (en) | Crowd's incident detection and appraisal procedure in a kind of video based on comentropy | |
CN101303726A (en) | System for tracking infrared human body target based on corpuscle dynamic sampling model | |
CN116310943B (en) | Method for sensing safety condition of workers | |
CN114842560B (en) | Computer vision-based construction site personnel dangerous behavior identification method | |
CN115171022A (en) | Method and system for detecting wearing of safety helmet in construction scene | |
CN116416577A (en) | Abnormality identification method for construction monitoring system | |
CN104866830A (en) | Abnormal motion detection method and device | |
Liu et al. | Abnormal crowd behavior detection based on optical flow and dynamic threshold | |
CN111274872B (en) | Video monitoring dynamic irregular multi-supervision area discrimination method based on template matching | |
CN115661755A (en) | Worker safety dynamic evaluation method based on computer vision and trajectory prediction | |
Zhao et al. | Abnormal behavior detection based on dynamic pedestrian centroid model: Case study on U-turn and fall-down |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |