CN109299700A - Subway group abnormal behavior detection method based on crowd density analysis - Google Patents
Subway group abnormal behavior detection method based on crowd density analysis Download PDFInfo
- Publication number
- CN109299700A CN109299700A CN201811195154.5A CN201811195154A CN109299700A CN 109299700 A CN109299700 A CN 109299700A CN 201811195154 A CN201811195154 A CN 201811195154A CN 109299700 A CN109299700 A CN 109299700A
- Authority
- CN
- China
- Prior art keywords
- crowd
- characteristic point
- group
- feature
- density
- 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
- 238000004458 analytical method Methods 0.000 title claims abstract description 23
- 238000001514 detection method Methods 0.000 title claims abstract description 15
- 206010000117 Abnormal behaviour Diseases 0.000 title abstract description 7
- 230000033001 locomotion Effects 0.000 claims abstract description 56
- 238000000034 method Methods 0.000 claims abstract description 43
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 17
- 230000002159 abnormal effect Effects 0.000 claims abstract description 14
- 230000003287 optical effect Effects 0.000 claims abstract description 13
- 238000005381 potential energy Methods 0.000 claims abstract description 13
- 230000008859 change Effects 0.000 claims abstract description 12
- 230000005856 abnormality Effects 0.000 claims description 18
- 230000003542 behavioural effect Effects 0.000 claims description 15
- 239000000284 extract Substances 0.000 claims description 13
- 230000001133 acceleration Effects 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 239000000203 mixture Substances 0.000 claims description 6
- 238000012937 correction Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000000717 retained effect Effects 0.000 claims description 2
- 230000006641 stabilisation Effects 0.000 claims description 2
- 238000011105 stabilization Methods 0.000 claims description 2
- 230000008030 elimination Effects 0.000 claims 1
- 238000003379 elimination reaction Methods 0.000 claims 1
- 238000012544 monitoring process Methods 0.000 abstract description 17
- 230000000694 effects Effects 0.000 abstract description 6
- 230000006399 behavior Effects 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 8
- 230000006872 improvement Effects 0.000 description 6
- 230000002547 anomalous effect Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000877 morphologic effect Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 206010016275 Fear Diseases 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000739 chaotic effect Effects 0.000 description 1
- 230000005284 excitation Effects 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000000505 pernicious effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000032258 transport Effects 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
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/53—Recognition of crowd images, e.g. recognition of crowd congestion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Social Psychology (AREA)
- Psychiatry (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a subway group abnormal behavior detection method based on crowd density analysis, which overcomes the problems of poor real-time performance, poor adaptability and single identified abnormal condition in the prior art, the abnormal behavior of the group is detected through the group density characteristic and the motion characteristic, the group density estimation algorithm is used for extracting the group density characteristic and the change rate of the number of people of the group, the sparse optical flow method is used for extracting two motion characteristics of group average kinetic energy and group motion direction entropy, the ORB characteristic point is used for extracting distance potential energy, the four group motion characteristics are combined with the density characteristic to realize the detection of the abnormal behavior, the detection accuracy under different group densities is improved, the method is suitable for different scenes, the effect is obvious, the effect is higher, and the identified abnormal conditions are more in types, so that the management and the layout of rail transit workers are facilitated, and the method is a great innovative breakthrough in the technical field of rail transit intelligent monitoring.
Description
Fields
The invention belongs to urban track traffic intelligent monitoring technology fields, and in particular to it is a kind of based on crowd density analysis
Subway group abnormality behavioral value method.
Background technique
In recent years, along with the rapid development of China's economic, rail traffic is saved with its land used, capacity is big, runing time is steady
The advantages such as fixed, safety gradually become each large- and-medium size cities and alleviate traffic pressure, increase the important channel of urban attraction,
It is the indispensable mode of transportation of numerous citizen's daily trip.At the same time, as the important component of rail traffic, subway
The safety problem at station but allows of no optimist.Subway station is complicated with scene, the features such as passenger flow mobility is big, density is high, these
All bring larger difficulty to security control, thus subway station become group security incident it is high-incidence, easily send out region.2017
On July 21, when No. 7 villages alignment Che Huanggang of Shenzhen Metro enter the station at station, a passenger is run because of unknown cause in compartment, is made one
Group, which gets a fright, causes fear, causes several passengers injured.The same year September 15 days, pa gloomy Green's subway station in London occurred quick-fried
Fried, several passengers are injured in tread event, and prime minister Te Leishamei announces for Britain's threat of terrorism rank to be adjusted to highest level " danger
It is anxious ".
Mass disturbance occurs in subway station can usually induce large-scale harm, these harm can not only destroy people
Health, can also bring huge public and private economic asset loss.When the Mass disturbance for influencing public safety occurs, people
Can instinct flee danger, at this moment caused by fear crowd's tread event is that have catastrophic, is typically due to people
It group crowded and trample and will lead to unexpected death by accident.Therefore, in subway monitoring system, how crowd's shape is effectively detected
State prevents crowded caused pernicious tread event, the bring loss for the accident having occurred and that is preferably minimized, these are all
It is problem urgently to be resolved.Although as riot etc. criminal offences be often happen suddenly, it is difficult to predict with avoid, by group
The accidents such as trampling caused by activity is that can prevent and mitigate by some perfect management means.Video monitoring system is not
Only the real-time video information for monitoring purpose can be provided for people, can also be used as a kind of subsequent tool for searching information.
So video monitoring system is in traffic, the industries such as public safety, bank as a kind of convenience and effective monitoring means
By universal use.At the same time, the development of numerous subject technologies such as image procossing, machine learning, artificial intelligence also ensures
Monitoring technology has popularity and brilliant scientific research and commercial value.
Traditional video monitoring system monitors numerous monitor videos merely by monitoring personnel, and then finds view
The anomalous event occurred in frequency can't possess the ability of anomalous event in intelligent and automation detection video.This for
Limit to very much for the demand of people.On the one hand, since available human resources are limited and due to the energy by people
Limitation, it is uninteresting for a long time to stare at monitor screen that people be allowed to generate is tired;On the other hand, the monitor video that video wall can be shown is certain
, the quantity for the camera that monitoring personnel effectively can be watched and be analyzed and time are also limited, and often result in miss very in this way
More important useful information.Therefore, the monitoring system in current practice cannot be analyzed well and processing colony is abnormal
Behavior, these numerous monitoring systems only serve as the tool of one " subsequent inquiry ".It is original therefore, it is necessary to improve
Video monitoring method.Currently, being checked using intelligent video monitoring system as the security and guard technology indirect labor of new generation of representative to identify
It is potential dangerous, become hot spot concerned by people.
Traditional unusual checking algorithm is there are real-time and robustness can not get both, rate of false alarm and rate of failing to report are larger, field
The problems such as abnormal conditions that scape adaptability is poor, identifies are single, these problems are all urgently to be resolved.How subway Scene realization is directed to
The optimization of detection algorithm, to subway station inlet and outlet, interior main region etc. of standing crowds' flowing is larger, crowd density is higher and easy
There is the region of abnormal conditions, monitors region in-group motion conditions in real time, analysis identifies group abnormality situation and reports in time
Police is a big difficulty.
Summary of the invention
The present invention is exactly directed to the problems of the prior art, and it is different to provide a kind of subway group based on crowd density analysis
Normal behavioral value method, overcomes that real-time in the prior art is poor, and adaptability is not strong, and the abnormal conditions of identification are single to ask
Topic, the detection of group abnormality behavior is carried out by population density feature and motion feature, use groups density estimation algorithm mentions
Population density feature and group's number change rate are taken, group's mean kinetic energy and group movement direction entropy are extracted using sparse optical flow method
Both motion features are realized four kind of groups motion feature combination density features different using ORB feature point extraction apart from potential energy
The detection of Chang Hangwei improves the detection accuracy under different crowd density, and suitable for different scenes, effect is obvious, and effect is more
Height, and the abnormal conditions type recognized is more, is intelligent track-traffic convenient for the management and layout of rail traffic staff
The big innovative breakthrough of the one of monitoring technology field.
To achieve the goals above, the technical solution adopted by the present invention is that: based on crowd density analysis subway group it is different
Normal behavioral value method, includes the following steps:
S1, frame image preprocessing: the pretreatment includes at least the conversion of color space, noise is eliminated, image enhancement and
Morphological scale-space;
S2 extracts ORB characteristic point:
S21 extracts oFAST characteristic point: carrying out pyramid sampling to pretreated frame image, is detected using FAST algorithm
The position of each tomographic image point of interest out recycles Harris characteristic point detection methods of marking sequence characteristic point obtained, choosing
Wherein N number of best point is taken, the direction of each characteristic point is calculated according to gray scale centroid method to N number of characteristic point of acquisition;
S22 extracts rBRIEF feature: BREIF feature is extracted, using the direction of oFAST characteristic point as the direction of BRIEF,
It is rotated, obtains steered BREIF, recycle greedy learning algorithm, find out 256 block of pixels to making its correlation most
It is low, and required Feature Descriptor is constituted, it filters out with high variance and high incoherent steered BEIEF, as
rBRIEF;
S23, comprehensive oFAST and rBRIEF describe to form ORB characteristic point.
S3, target feature point is extracted and tracking: calculating the light at characteristic point using pyramid Lucas-Kanade optical flow method
Stream obtains moving target information;
Population density analysis: S4 extracts ROI (Region Of Interest) using improved mixture Gaussian background model
Two-value foreground area is corrected by gridding method and updated to interior two-value foreground area, and using the setting of ORB feature dot density, group is close
Weighted value is spent, foreground pixel area normalization is completed, respective least square method matched curve is used within the scope of different weights
Population density is estimated, and density is classified;
S5, crowd movement's feature extraction: crowd movement's feature include: crowd's mean kinetic energy, crowd movement direction entropy,
Apart from potential energy and individual average acceleration between individual in crowd.
S6, unusual checking:
S61 excludes low-density and Dense crowd situation according to the population density that step S4 is obtained;
Crowd movement's feature that S62, analytical procedure S5 are extracted is realized and classifies to normal or abnormal situation.
It as an improvement of the present invention, further include the grid for dividing an image into different size grid in the step S1
Method alleviates perspective distortion, and the size of the grid is inversely proportional with the pixel region apart from video camera distance, and assigns corresponding power
It is worth, the pixel characteristic after statistical weight.
It is improved as another kind of the invention, the step S21 extracts oFAST characteristic point and further comprises: defining any
The neighborhood territory pixel square of one characteristic point I (x, y) pixel are as follows:
So center-of-mass coordinate of image are as follows:
Then, the angle of characteristic point and mass center is the direction of FAST characteristic point:
It is improved as another kind of the invention, the step S22 extracts rBRIEF feature and further comprises: with each feature
Centered on point:
Wherein, p (x) is the gray value at x, then BRIEF description are as follows:
By n point to one matrix of composition:
Utilize field direction θ and corresponding spin matrix Rθ, construct a correction version S of Sθ;
Sθ=RθS
In formula,It then can get steered BRIEF description are as follows:
gn(p, θ)=fn(p)|(xi,yi)∈Sθ。
As another improvement of the invention, the step S3 further comprises:
S31 judges the quantity of current ORB characteristic point, if quantity is less than setting value, return step S2 is extracted again, otherwise
Continue;
S32 is tracked ORB characteristic point using pyramid Lucas-Kanade optical flow method, estimates current frame image phase
The position of each corresponding characteristic point, if the brightness value of characteristic point in allowable error, is tracked successfully and retained, otherwise remove
It goes;
S33, the excessive or too small noise characteristic point of removal movement position, the moving distance of every a pair of of the characteristic point of two frame of front and back
As feature-point optical flow corresponding to present frame;
S34, using next frame image as current frame image, return step S2 continues to extract ORB characteristic point or directly carry out
Tracking, until image trace finishes.
It is improved as another of the invention, in the step S4, crowd's quantity is calculated using two-value foreground area
Change rate.
As a further improvement of the present invention, crowd's mean kinetic energy E in the step S5avgAre as follows:
Wherein, EkIt is the sum of crowd movement's energy, is indicated by calculating the light stream energy in area-of-interest;N is people
Group's density;
Crowd movement direction entropy includes light stream vector direction histogram, Direction Probability distribution map and direction entropy.
As a further improvement of the present invention, in the step S62, crowd's amount change reflects crowd's quantity
The speed of variation, the stabilization of numerical value whether, realize that normal or abnormal situation determines;
The speed degree and severe degree of crowd's mean kinetic energy reflection group movement speed, judge whether crowd fears
Unbearably;
Crowd movement direction entropy characterizes the confusion degree in crowd movement direction, and numerical value is bigger, indicates crowd movement side
To exception, occur equidirectional prominent scattered;
Degree of scatter in the crowd between individual between potential energy reflection group two-by-two individual, if uprushing apart from potential energy
Or anticlimax, illustrate there is a possibility that abnormal conditions generation;
The severity of the individual average acceleration reflection crowd movement, numerical value is higher, and abnormality is bigger
Compared with prior art, the invention proposes a kind of subway group abnormality behavioral values based on crowd density analysis
Method, the advantage is that:
(1), higher for real-time, rapidity requirement in subway station security protection work, introduce ORB (oriented FAST
And rotated BRIEF) feature extraction algorithm, ORB algorithm is 100 times of SIFT algorithm in calculating speed, is that SURF is calculated
10 times of method, meet actual demand.
(2), it is based on ORB feature extraction algorithm, using the sparse optical flow method based on ORB characteristic point, to pyramid Lucas-
Kanade optical flow method is improved.
(3), a kind of population density algorithm for estimating conduct auxiliary is introduced, the asynchronous situation of crowd density is carried out at classification
Reason reduces erroneous judgement, improves working efficiency.
(4), the population density algorithm for estimating calculating proposed is simple, judges that speed is fast, discrimination is higher.
(5), it reduces while ensureing real-time and is judged by accident under low-density scene, improve the attention to high-density scene
Degree.
(6), lower for camera in subway station, distance of the moving target apart from camera is obstructed to wait brought perspective
Aberration problems propose a kind of group abnormality behavioral value method of band correction, correct moving target ruler in scene by gridding method
Dynamic change inconsistency caused by very little variation, establishes adaptive code of conduct, reaches increase monitoring range, reduces false detection rate
The purpose of.
Detailed description of the invention
Fig. 1 is detection method flow diagram of the invention;
Fig. 2 is target feature point detection and trace flow schematic diagram.
Specific embodiment
Below with reference to drawings and examples, the present invention is described in detail.
Embodiment 1
Based on the subway group abnormality behavioral value method of crowd density analysis, as shown in Figure 1, including the following steps:
Frame image preprocessing: S1 carries out the pretreatment of related algorithm to each frame image that camera is shot, solves
Image is because of the problems such as image caused by the reasons such as environment or shooting orientation angles is obscured, deformed.It mainly include color space
Conversion, Morphological scale-space, Image Denoising Technology, image Shadows Processing technology, image enhancement technique, it is therefore intended that realize reduce
Picture noise, removal picture shade, setting area-of-interest, image enhancement, smooth and sharpening etc., improve subsequent population behavior inspection
The accuracy of survey reduces error.
Preprocessed video data set needs enough sample sizes, is able to reflect rules more as far as possible, and it is insufficient to reduce sample
Bring random error.Select UMN data set, PETS2009 data set, Line of Nanjing Subway monitor video as initial data here
Collection.
Gridding method is also introduced in step S1, video image is divided into different size of grid, the size of grid and the picture
The distance of plain region distance video camera is inversely proportional, and assigns corresponding weight, and the pixel characteristic after statistical weight can be certain
Alleviate perspective distortion in degree.
S2 extracts ORB characteristic point:
S21 extracts oFAST characteristic point: carrying out pyramid sampling to pretreated frame image, is detected using FAST algorithm
The position of each tomographic image point of interest out recycles Harris characteristic point detection methods of marking sequence characteristic point obtained, choosing
Wherein N number of best point is taken, the direction of each characteristic point is calculated according to gray scale centroid method to N number of characteristic point of acquisition, it is fixed
The neighborhood territory pixel square of any one characteristic point I (x, y) pixel of justice are as follows:
So center-of-mass coordinate of image are as follows:
The angle of feature points and mass center is the direction of FAST characteristic point:
Thus oFAST (Orientation FAST) characteristic point is obtained;
S22 extracts rBRIEF feature: BREIF feature is extracted, using the direction of oFAST characteristic point as the direction of BRIEF,
It is rotated, obtains steered BREIF, recycle greedy learning algorithm, find out 256 block of pixels to making its correlation most
It is low, and required Feature Descriptor is constituted, it filters out with high variance and high incoherent steered BEIEF, as
RBRIEF, centered on each characteristic point, definition:
Wherein, p (x) is the gray value at x, and BRIEF description is the two-value sequence that a length is n, by characteristic point week
N point is enclosed to generation.Then BRIEF description are as follows:
By n point to one matrix of composition:
Utilize field direction θ and corresponding spin matrix Rθ, construct a correction version S of Sθ;
Sθ=RθS
In formula,It then can get steered BRIEF description are as follows:
gn(p, θ)=fn(p)|(xi,yi)∈Sθ;
S23, comprehensive oFAST and rBRIEF describe to form ORB characteristic point.
S3, target feature point is extracted and tracking: calculating the light at characteristic point using pyramid Lucas-Kanade optical flow method
Stream obtains moving target information, as shown in Figure 2;
S31 judges the quantity of current ORB characteristic point, if quantity is less than setting value, is not enough to characterize the movement of individual goal
Behavior region needs to extract again in adjacent next frame, and until meeting the requirements, thus return step S2 is extracted again, otherwise
Continue;
S32 is tracked ORB characteristic point using pyramid Lucas-Kanade optical flow method, in next frame fixed size
Window in corresponding each characteristic point of estimation current frame image position, if the brightness value of characteristic point is in allowable error
It is interior, then it tracks successfully and retains, otherwise remove;
S33, the excessive or too small noise characteristic point of removal movement position, the moving distance of every a pair of of the characteristic point of two frame of front and back
As feature-point optical flow corresponding to present frame;
S34, using next frame image as current frame image, return step S2 continues to extract ORB characteristic point or directly carry out
Tracking, until image trace finishes.
Population density analysis: S4 extracts ROI (Region Of Interest) using improved mixture Gaussian background model
Two-value foreground area is corrected by gridding method and updated to interior two-value foreground area, and using the setting of ORB feature dot density, group is close
Weighted value is spent, foreground pixel area normalization is completed, respective least square method matched curve is used within the scope of different weights
Population density is estimated, and density is classified, when population density is lower than minimum predetermined threshold, is considered as the safe feelings of group
Condition, to reduce False Rate;When population density, which is higher than highest, presets threshold, it is considered as group abnormality situation.Finally using before two-value
Scape areal calculation obtains crowd's amount change.
S5, crowd movement's feature extraction: the mainly individual spacing from crowd's mean kinetic energy, crowd movement direction entropy, crowd
Crowd movement's feature is characterized from this four potential energy, individual average acceleration aspects.Wherein, by calculating in area-of-interest
Light stream energy indicates the sum of crowd movement's energy Ek, due to EkIt is related with crowd density N, define crowd's mean kinetic energy EavgCarry out table
The severity of group movement is levied, wherein
The confusion degree of group's direction of motion is characterized by crowd movement direction entropy, it is mainly straight comprising light stream vector direction
Fang Tu, Direction Probability distribution map and direction entropy;Point of crowd is characterized by the distance between individual each in crowd potential energy
Cloth situation;It is described by individual average acceleration when anomalous event occurs to apply itself since crowd is run suddenly
Active force increase leads to crowd behaviour varying motion.
It is extracted five kinds of crowd characteristics in S6, unusual checking, step S4 and step S5 altogether: crowd's amount change,
Crowd's mean kinetic energy, crowd movement direction entropy, in crowd between individual under potential energy, individual average acceleration, various situations, often
A kind of crowd behaviour has the characteristics that respective and different changing features, so by random gloomy made of training sample training
Woods classifies to crowd behaviour by support vector machines (SVM).
S61 excludes low-density and Dense crowd situation according to the population density that step S4 is obtained, and reduces erroneous judgement simultaneously
Promote the supervision to crowd's situation under high density case;
Crowd movement's feature that S62, analytical procedure S5 are extracted is realized and classifies to normal or abnormal situation:
Crowd's amount change reflects the speed of crowd's quantity variation, if crowd's no exceptions behavior, video capture
Number change rate in range should be relatively stable, and when certain abnormal behaviours occur, number change rate can uprush phenomenon;
Crowd's mean kinetic energy characterizes the speed speed degree and severe degree of group movement, can be used to judge the walking states of group
It is normal walking or generally runs;Crowd movement direction entropy characterizes the confusion degree in crowd movement direction, crowd in subway station
Different in the direction of motion of different situations, in regional movements such as station layers, the direction of motion is more chaotic always, if group transports
Dynamic direction becomes very single suddenly, there is the probability for the behavior of being abnormal, and in regions such as station hall layers, the direction of motion of crowd is answered
This is more unified, if direction of motion entropy increases at this time, illustrates crowd movement direction exception;Apart from potential energy between individual in crowd
Characterize group's the distance between individual situation two-by-two, the degree of scatter between description group individual, if uprushing or dashing forward apart from potential energy
Subtract, then illustrate that interpersonal distance changes suddenly, explanation has a possibility that abnormal behaviour generation at this time;Individual is average
Accelerometer levies the severity of crowd movement, and when facing a danger situation, crowd's instinct is fled danger, and people itself, which has an effect, to be added
Greatly, excitation people accelerates to run, and the severity of movement obviously increases.Realize herein to normal, crowd panic, it is equidirectional it is prominent dissipate,
Whip scurry, conflict several situations of group realize classification.
The basic principles, main features and advantages of the present invention have been shown and described above.The technology of the industry
Personnel only illustrate the present invention it should be appreciated that the present invention is not limited by examples detailed above described in examples detailed above and specification
Principle, various changes and improvements may be made to the invention without departing from the spirit and scope of the present invention, these variation and
Improvement all fall within the protetion scope of the claimed invention.The claimed scope of the invention is by appended claims and its is equal
Object defines.
Claims (8)
1. the subway group abnormality behavioral value method based on crowd density analysis, which comprises the steps of:
S1, frame image preprocessing: the pretreatment includes at least the conversion of color space, noise elimination, image enhancement and form
Processing;
S2 extracts ORB characteristic point:
S21 extracts oFAST characteristic point: carrying out pyramid sampling to pretreated frame image, is detected often using FAST algorithm
The position of one tomographic image point of interest recycles Harris characteristic point detection methods of marking sequence characteristic point obtained, chooses it
In N number of best point, the direction of each characteristic point is calculated according to gray scale centroid method to N number of characteristic point of acquisition;
S22 extracts rBRIEF feature: extracting BREIF feature, using the direction of oFAST characteristic point as the direction of BRIEF, carries out
Rotation, obtains steered BREIF, recycles greedy learning algorithm, finds out 256 block of pixels to keeping its correlation minimum, and
Feature Descriptor needed for constituting is filtered out with high variance and high incoherent steered BEIEF, as rBRIEF;
S23, comprehensive oFAST and rBRIEF describe to form ORB characteristic point.
S3, target feature point is extracted and tracking: calculating the light stream at characteristic point using pyramid Lucas-Kanade optical flow method, obtains
Take moving target information;
Population density analysis: S4 is extracted in ROI (Region Of Interest) using improved mixture Gaussian background model
Two-value foreground area is corrected by gridding method and updated to two-value foreground area, is weighed using ORB feature dot density setting population density
Weight values complete foreground pixel area normalization, estimate within the scope of different weights with respective least square method matched curve
Population density out, and density is classified;
S5, crowd movement's feature extraction: crowd movement's feature includes: crowd's mean kinetic energy, crowd movement direction entropy, crowd
Apart from potential energy and individual average acceleration between middle individual.
S6, unusual checking:
S61 excludes low-density and Dense crowd situation according to the population density that step S4 is obtained;
Crowd movement's feature that S62, analytical procedure S5 are extracted is realized and classifies to normal or abnormal situation.
2. the subway group abnormality behavioral value method as described in claim 1 based on crowd density analysis, it is characterised in that
Further include dividing an image into the gridding method of different size grid in the step S1 to alleviate perspective distortion, the grid it is big
It is small to be inversely proportional with the pixel region apart from video camera distance, and corresponding weight value is assigned, the pixel characteristic after statistical weight.
3. the subway group abnormality behavioral value method as claimed in claim 1 or 2 based on crowd density analysis, feature exist
Extracting oFAST characteristic point in the step S21 further comprises: any one characteristic point I (x, y) pixel neighborhood of a point picture of definition
Plain square are as follows:
So center-of-mass coordinate of image are as follows:
Then, the angle of characteristic point and mass center is the direction of FAST characteristic point:
4. the subway group abnormality behavioral value method as claimed in claim 1 or 2 based on crowd density analysis, feature exist
Extracting rBRIEF feature in the step S22 further comprises: centered on each characteristic point:
Wherein, p (x) is the gray value at x, then BRIEF description are as follows:
By n point to one matrix of composition:
Utilize field direction θ and corresponding spin matrix Rθ, construct a correction version S of Sθ;
Sθ=RθS
In formula,It then can get steered BRIEF description are as follows:
gn(p, θ)=fn(p)|(xi,yi)∈Sθ。
5. the subway group abnormality behavioral value method as described in claim 1 based on crowd density analysis, it is characterised in that
The step S3 further comprises:
S31 judges the quantity of current ORB characteristic point, if quantity is less than setting value, return step S2 is extracted again, otherwise continued;
S32 is tracked ORB characteristic point using pyramid Lucas-Kanade optical flow method, and estimation current frame image is corresponding
Each characteristic point position, if the brightness value of characteristic point in allowable error, is tracked successfully and is retained, otherwise remove;
S33, removes the excessive or too small noise characteristic point of movement position, and the moving distance of every a pair of of the characteristic point of two frame of front and back is
Feature-point optical flow corresponding to present frame;
S34, using next frame image as current frame image, return step S2 continues to extract ORB characteristic point or directly be tracked,
Until image trace finishes.
6. the subway group abnormality behavioral value method as described in claim 1 based on crowd density analysis, it is characterised in that
In the step S4, crowd's amount change is calculated using two-value foreground area.
7. the subway group abnormality behavioral value method according to claim 1 based on crowd density analysis, feature exist
Crowd's mean kinetic energy E in the step S5avgAre as follows:
Wherein, EkIt is the sum of crowd movement's energy, is indicated by calculating the light stream energy in area-of-interest;N is that crowd is close
Degree;
Crowd movement direction entropy includes light stream vector direction histogram, Direction Probability distribution map and direction entropy.
8. the subway group abnormality behavioral value method according to claim 6 based on crowd density analysis, feature exist
In the step S62, the speed of crowd's amount change reflection crowd's quantity variation, the stabilization of numerical value whether, is realized
Normal or abnormal situation determines;
The speed degree and severe degree of crowd's mean kinetic energy reflection group movement speed, judge whether crowd is panic;
Crowd movement direction entropy characterizes the confusion degree in crowd movement direction, and numerical value is bigger, indicates that crowd movement direction is different
Often, occur equidirectional prominent scattered;
Degree of scatter in the crowd between individual between potential energy reflection group two-by-two individual, if uprushing or dashing forward apart from potential energy
Subtract, illustrates there is a possibility that abnormal conditions generation;
The severity of the individual average acceleration reflection crowd movement, numerical value is higher, and abnormality is bigger.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811195154.5A CN109299700A (en) | 2018-10-15 | 2018-10-15 | Subway group abnormal behavior detection method based on crowd density analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811195154.5A CN109299700A (en) | 2018-10-15 | 2018-10-15 | Subway group abnormal behavior detection method based on crowd density analysis |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109299700A true CN109299700A (en) | 2019-02-01 |
Family
ID=65162610
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811195154.5A Pending CN109299700A (en) | 2018-10-15 | 2018-10-15 | Subway group abnormal behavior detection method based on crowd density analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109299700A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110263690A (en) * | 2019-06-12 | 2019-09-20 | 成都信息工程大学 | A kind of group behavior feature extraction based on small group and description method and system |
CN110781723A (en) * | 2019-09-05 | 2020-02-11 | 杭州视鑫科技有限公司 | Group abnormal behavior identification method |
CN111753651A (en) * | 2020-05-14 | 2020-10-09 | 南京熊猫电子股份有限公司 | Subway group abnormal behavior detection method based on station two-dimensional crowd density analysis |
CN111914594A (en) * | 2019-05-08 | 2020-11-10 | 四川大学 | Group emotion recognition method based on motion characteristics |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104680140A (en) * | 2015-02-13 | 2015-06-03 | 电子科技大学 | Image-based crowd massing state detection method |
CN104732236A (en) * | 2015-03-23 | 2015-06-24 | 中国民航大学 | Intelligent crowd abnormal behavior detection method based on hierarchical processing |
-
2018
- 2018-10-15 CN CN201811195154.5A patent/CN109299700A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104680140A (en) * | 2015-02-13 | 2015-06-03 | 电子科技大学 | Image-based crowd massing state detection method |
CN104732236A (en) * | 2015-03-23 | 2015-06-24 | 中国民航大学 | Intelligent crowd abnormal behavior detection method based on hierarchical processing |
Non-Patent Citations (2)
Title |
---|
宋丹妮: "基于视频监控的群体异常行为检测研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
梁玉: "基于ORB 兴趣点的异常行为检测技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111914594A (en) * | 2019-05-08 | 2020-11-10 | 四川大学 | Group emotion recognition method based on motion characteristics |
CN111914594B (en) * | 2019-05-08 | 2022-07-01 | 四川大学 | Group emotion recognition method based on motion characteristics |
CN110263690A (en) * | 2019-06-12 | 2019-09-20 | 成都信息工程大学 | A kind of group behavior feature extraction based on small group and description method and system |
CN110781723A (en) * | 2019-09-05 | 2020-02-11 | 杭州视鑫科技有限公司 | Group abnormal behavior identification method |
CN110781723B (en) * | 2019-09-05 | 2022-09-02 | 杭州视鑫科技有限公司 | Group abnormal behavior identification method |
CN111753651A (en) * | 2020-05-14 | 2020-10-09 | 南京熊猫电子股份有限公司 | Subway group abnormal behavior detection method based on station two-dimensional crowd density analysis |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
CN111160125B (en) | Railway foreign matter intrusion detection method based on railway monitoring | |
CN103942811B (en) | Distributed parallel determines the method and system of characteristic target movement locus | |
CN103839065B (en) | Extraction method for dynamic crowd gathering characteristics | |
CN106128053A (en) | A kind of wisdom gold eyeball identification personnel stay hover alarm method and device | |
CN109299700A (en) | Subway group abnormal behavior detection method based on crowd density analysis | |
Kong et al. | Detecting abandoned objects with a moving camera | |
CN100583128C (en) | Real time intelligent control method based on natural video frequency | |
CN106778655B (en) | Human body skeleton-based entrance trailing entry detection method | |
CN102521565A (en) | Garment identification method and system for low-resolution video | |
CN105574506A (en) | Intelligent face tracking system and method based on depth learning and large-scale clustering | |
CN111753651A (en) | Subway group abnormal behavior detection method based on station two-dimensional crowd density analysis | |
CN107659754B (en) | Effective concentration method for monitoring video under condition of tree leaf disturbance | |
CN105426820A (en) | Multi-person abnormal behavior detection method based on security monitoring video data | |
CN106127814A (en) | A kind of wisdom gold eyeball identification gathering of people is fought alarm method and device | |
CN104616006A (en) | Surveillance video oriented bearded face detection method | |
CN106033548B (en) | Crowd abnormity detection method based on improved dictionary learning | |
Ullah et al. | Gaussian mixtures for anomaly detection in crowded scenes | |
CN111091057A (en) | Information processing method and device and computer readable storage medium | |
Ghidoni et al. | Texture-based crowd detection and localisation | |
KR20140132140A (en) | Method and apparatus for video surveillance based on detecting abnormal behavior using extraction of trajectories from crowd in images | |
Li et al. | End-to-end multiplayer violence detection based on deep 3D CNN | |
Zhao et al. | Pedestrian motion tracking and crowd abnormal behavior detection based on intelligent video surveillance | |
Maheshwari et al. | A review on crowd behavior analysis methods for video surveillance | |
CN103646420A (en) | Intelligent 3D scene reduction method based on self learning algorithm |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190201 |
|
RJ01 | Rejection of invention patent application after publication |