CN103559478B - Overlook the passenger flow counting and affair analytical method in pedestrian's video monitoring - Google Patents
Overlook the passenger flow counting and affair analytical method in pedestrian's video monitoring Download PDFInfo
- Publication number
- CN103559478B CN103559478B CN201310459104.4A CN201310459104A CN103559478B CN 103559478 B CN103559478 B CN 103559478B CN 201310459104 A CN201310459104 A CN 201310459104A CN 103559478 B CN103559478 B CN 103559478B
- Authority
- CN
- China
- Prior art keywords
- target
- feature
- rectangular
- pedestrian
- people
- 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.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention proposes the passenger flow countings and affair analytical method in a kind of vertical view pedestrian's video monitoring.Main includes the feature extraction and detection for overlooking pedestrian, overlooks the tracking of pedestrian and overlooks the analysis etc. of pedestrian's monitor event.The detection for overlooking pedestrian is based on overlooking pedestrian head feature, and head feature can resolve into several cascade feature combinations, be successively: rectangular characteristic, colour stable feature, characteristics of gradient change, trapezoidal characteristics and circular arc feature etc.;The tracking for overlooking pedestrian uses adaptive Mean-shift algorithm, and multiple target may be implemented and obtain and track automatically;Overlook the classification and discriminating conduct of passenger flow event.
Description
Technical field
The present invention relates to the intelligent Video Surveillance Technologies in computer vision field.
Background technique
The present invention is suitable for communal facility entrance and passageway, and photographic device is placed in top, overlooks shooting downwards.Such as public affairs
Hand over vehicle entrance, supermarket's toll collection lanes, Entrance passageway etc., 2 meters away from ground of camera or more, be mainly used for passenger flow statistics and
The monitoring of the stream of people.It is relatively fewer to the detection for overlooking passenger flow and enumeration problem research both at home and abroad, especially at home, connect view at present
Basic passenger flow counting function does not have or not perfect in frequency.
Summary of the invention
There are two the task of vertical view pedestrian's video monitoring is main: one is passenger flow counting, and one is event analysis.Passenger flow meter
Several contents is mainly detection, tracking and the counting under general work state to passenger flow is overlooked.The event point for overlooking passenger flow is common
Event and two kinds of anomalous event;The common event have wait in line, personnel's congestion etc., anomalous event has scene to leave article and two people
It fights.
(1) detection of pedestrian is overlooked
1) triggering of monitoring system
It is triggered using moving image detection, i.e., calculates the variation of scene content using video frame differences method, setting video is adjacent
The gray level image of frame subtracts each other to obtain frame difference figure Imn, i.e. Imn=Im(n)-Im(n-1)
In formula, n indicates that the ordinal number of present frame, Im (n) indicate that present frame gray image, Im (n-1) indicate previous frame gray scale
Image;It is changed if any present frame and previous frame content and changes to meet certain requirements and (such as detected super with previous frame variable quantity
It crosses 1%), just the detection of pedestrian is overlooked in triggering.
2) target is detected using graded features
It is detection target that vertical view pedestrian detection of the invention, which is to overlook the number of people of pedestrian,.When detection system be triggered with
Afterwards, the just traversal detection target on current frame image, i.e., with a certain size hough transform frame with a certain size step-length from upper
Under, whole image is scanned from left to right.The content of each detection rectangle frame is candidate target.The feature of candidate target includes
Rectangular characteristic, colour stable feature, characteristics of gradient change, trapezoidal characteristics and circular arc feature etc..
Following characteristics detection is cascade structure, and the target rectangle frame for only meeting preceding features just can be carried out next link
Feature detection, the candidate target rectangle frame for being unsatisfactory for condition are abandoned in the detection for being unsatisfactory for feature.
1. rectangular characteristic detects
The rectangular characteristic of target is mainly between the rectangular block of target part based on difference, i.e. the characteristic of target appearance
The difference of the gray scale accumulated value (abbreviation rectangle value can quickly be calculated with " integrogram " and be obtained) of each rectangular partition pixel can be used
Value judges.Difference between rectangular partition reflects the difference of two pieces of rectangular appearances, if setting the threshold value of relative deviation,
The whole or local symmetrical feature of the target can be determined according to the size of relative deviation between different rectangular blocks.By target
It is regarded as rectangle, target signature is according to the rectangular partition correlation and characteristic distributions decision in the rectangle.Number of people target is in rectangle
In feature, the features such as symmetrical the right and left, upper and lower two half similar, three partitioned organizations and lumpiness are used, as shown in Figure 1.
Specific descriptions are whether symmetrically to detect target rectangle frame, i.e. whether the two sub- rectangular blocks in left and right are not much different;Detect target
The whether upper lower aprons of rectangle frame, i.e., whether two sub- rectangular blocks are not much different up and down, but must have certain difference;Detect mesh
Mark rectangle frame whether left, center, right structure, i.e., the sub- rectangular block on intermediate sub- rectangular block and both sides meets certain difference requirements;Inspection
Survey whether target rectangle frame has an agglomerate feature, i.e., at least there are three be different from intermediate sub- rectangular block in the sub- rectangular block at four angles
Value.
2. color stability characteristics
Head part's color is one metastable agglomerate of rectangular centre, so calculating rectangular centre pixel gray value
Variance meets the feature of the number of people, does not otherwise just meet if variance under threshold value, is regarded as the target rectangle block colour stable.
3. characteristics of gradient change (representation of concept for borrowing " gradient " here has a greater change)
Eigen refers to head zone color and background color should have a biggish difference, i.e., the color other than contouring head with
The color on head is different.The average value of rotation sampling background dot is compared with the average value of the number of people.As shown in Fig. 2, setting target pros
The centre coordinate of shape frame is (i, j), and target frame side length is 2R, and number of people sample radius r(r is less than number of people radius), from center of circle injection
Line rotates counterclockwise, θ indicate ray and horizontal position angle), sampling location be ray at θ=2pi/N head with exterior domain
That is the sampling of background.In number of people region, sampled point outside number of people region pixel value respectively indicates are as follows: and H(i+rcos θ, j+
Rsin θ) and B(i+Rcos θ, j+Rsin θ), then the change of gradient value calculating formula of the target are as follows:
Simplest sampling is only taken and is detected with the sampled point in 4 directions in upper and lower, left and right to calculate characteristics of gradient change.
Only just meet the requirement of target more than the result of certain threshold value.
4. trapezoidal characteristics
Human body head is a similar circular agglomerate under certain depression angle, and agglomerate is characterized in that internal color is equal
Even, edge is usually arc-shaped, then the arc-shaped shape feature of head edge, can be utilized as judging candidate target
Whether be human body head foundation.Circular arc can be with trapezoidal come approximate to simplify the calculation.There are two types of methods for trapezoidal characteristics detection.
A kind of trapezoidal characteristics in approach application rectangular characteristic are judged, as shown in figure 3, setting trapezoidal top (short side)
Outside, below (long side) inner, the trapezoidal rectangular block for occupying continuous two row or two column, and trapezoidal not in middle position then exists
The trapezoidal characteristic of four positions in upper and lower, left and right calculates judgement by following formula.
Another method, as shown in figure 4, this is the vertical view people for passing through image preprocessing, binaryzation, morphology operations
Head image, head zone show as complete white part in figure.Sequentially traversal is carried out to each row of target window to obtain continuously
White line segment, so pressing line head portion region, median is continuously that the line segment of 1 pixel composition indicates head in a line
In the length of the row.It is generally shorter closer to the line segment length of frame since contouring head is overall or major part is in arc-shaped, more
Longer by paracentral line segment, such line segment simultaneously can just have the effect of arc-shaped together.The length of AB line segment is less than in figure
The length of CD line segment, and so on, it begins to generate camber line;Similar, longest width EF is obtained, the line of next line is continued to scan on
Section GH, length reduce (AB and CD here, the line number between EF and GH are sequentially incremented by successively) instead.Left and right two halves
Calculating judges similar.In rectangular target frame, upper half, lower half, a left side half, right half part meet ladder as long as having a part to meet
Shape characteristic requirements.
5. circular arc/circle feature
Although the agglomerate feature that the number of people is presented, can use Hough transform, carries out when setting suitable parameter
Loop truss (practical is exactly circular-arc detection) is to reach actual needs.So asking edge and binaryzation to image in target rectangle frame
Processing, having detected whether that circular wheel profile judges whether is target.
(2) target following
After target is detected in specified region, with regard to carrying out the tracking of target, tracking characteristics select the gray scale of rectangle frame
Property of the histogram.The present invention is tracked using continuous adaptive mean shift (CAMShift) algorithm, and tracking target obtains automatically
It takes, and energy multiple target tracks simultaneously, algorithm key step is as follows:
1. selecting the initial position of search window, search window color probability distribution is calculated, initial position is by above-mentioned detection side
The target rectangle frame that method detects;
2. new window position is arranged with modified method, mean shift (Mean Shift) tracking is executed;
3. saving zeroth order square;
4. the size according to zeroth order away from setting search window;
5. repeat step 2., 4. until convergence.
(3) passenger flow counting and event analysis
1. detection of passenger flow counts: setting range ABCD in the picture, as shown in figure 5, when target is moved through CD line will be by
Detection, and the target is tracked, just stop tracking until crossing another side AB line, count is incremented;If there is target opposite direction is (first from AB
Line removed CD line) it is mobile, then it counts and subtracts 1;
2. people line: multiple target continues to move away in detection range, and slowly.
3. personnel's congestion: intensive multiple target is continued for some time not to be left in detection range.
4. residue: retaining original background, when motion detection does not change, and background continues difference and decides that scene at this time
There is residue.
5. two people fight: two target Continuous move in detection range.
Detailed description of the invention
The rectangular characteristic schematic diagram of Fig. 1 vertical view pedestrian head: the first row overlooked the original gradation figure of pedestrian head before this,
It and then is successively the uniform agglomerate of center color in rectangular characteristic, the second row is that the right and left is symmetrical, upper and lower two half similar,
The third line is three partitioned organizations (intermediate different from two sides) rectangular characteristic, and fourth line is agglomerate rectangular characteristic (intermediate rectangular block
It is all different with the rectangular block at four angles).
Fig. 2 includes the rectangle frame of number of people target.
16 piecemeal schematic diagrames of Fig. 3 rectangular characteristic, different sub-block groups can synthesize different features.
Number of people target after Fig. 4 binary conversion treatment.
Fig. 5 pedestrian is mobile and counts example.
Claims (6)
1. a kind of method of the passenger flow counting overlooked in pedestrian's video monitoring and event analysis: being to examine to the detection for overlooking pedestrian
Survey overlook pedestrian number of people feature based on, overlook pedestrian number of people feature include rectangular partition symmetrically with asymmetric feature, ladder
Shape feature, characteristics of gradient change, colour stable feature and arc section feature, these are characterized in combining in a manner of cascade classifier
Together, every first-level class device represents a feature decision, only meets the candidate target ability of this grade of classifier characteristic condition
Judge into next stage classifier;Human body head is that the similar round, internal color in an edge is uniformly rolled into a ball under depression angle
Block, for promoted detection efficiency, with it is trapezoidal calculating come it is approximate, there are two types of method;A kind of ladder in approach application rectangular partition feature
The judgement of shape feature;Another method is by after image preprocessing, binaryzation, morphology operations, by head zone in figure with complete
White indicates, traverses to every a line of target window by pixel, the white line segment length in horizontal direction is obtained, by one
The intermediate line segment continuously formed for the pixel of " 1 " of row indicates head in the length of the row, since contouring head is totally in circular arc
Shape, it is desirable that the line segment length closer to target window frame is shorter, and the line segment closer to target's center is longer, and such line segment is simultaneously
Can just have the effect of profile together in circular arc;By image number of people target be respectively classified into up and down and left and right two halves, i.e., upper half, under
Half, left half, right half, meet the requirement of target trapezoidal characteristic in this four part as long as thering is a part to meet.
2. according to the method described in claim 1, the rectangular partition of the target number of people symmetrically with asymmetric feature, it is characterized in that
Between the rectangular block of target part based on difference, i.e., the feature of target appearance can be tired with the gray scale of each rectangular partition pixel
Value added, abbreviation rectangle value judges with " integrogram " calculated difference;Difference between rectangular partition reflects two pieces
The difference of rectangular appearance, if setting the threshold value of deviation, so that it may which, according to the size of deviation between different rectangular blocks, determining should
The whole and local symmetrical feature of target;Target is regarded as rectangle, target signature is mutually closed according to the rectangular partition in the rectangle
System and characteristic distributions determine;For number of people target in rectangular characteristic, rectangular characteristic specific descriptions are whether detection target rectangle frame is left
Right symmetrical, i.e. whether the sub- rectangular block difference in left and right two is little;It is whether approximate up and down to detect target rectangle frame, i.e., upper and lower two sons
Whether rectangular block difference is little, but must have certain difference;Detect target rectangle frame whether left, center, right structure, i.e., it is intermediate
Sub- rectangular block and the sub- rectangular block on both sides meet certain difference requirements;Whether detection target rectangle frame has agglomerate feature, i.e.,
At least there are three the values for being different from intermediate sub- rectangular block in the sub- rectangular block at four angles.
3. according to the method described in claim 1, the color stability characteristics of the number of people, it is characterised in that: head part's face
Color is one metastable agglomerate of rectangular centre, so the variance of rectangular centre pixel gray value is calculated, if variance exists
Under threshold value, it is regarded as the target rectangle block colour stable, meets the feature of the number of people, does not otherwise just meet.
4. according to the method described in claim 1, the characteristics of gradient change of target is head zone overall color and background color
There should be biggish difference;The average value of the background dot of rotary shaft up-sampling is compared with the average value of the target number of people, only
More than the requirement that the result of certain threshold value just meets target.
5. according to the method described in claim 1, when target specified region be detected after, with regard to progress target tracking, with
The grey level histogram characteristic of track feature selecting rectangle frame;The present invention using continuous adaptive mean shift (CAMShift) algorithm into
Line trace, tracking target obtain automatically, and energy multiple target real-time tracking, algorithm steps are as follows: 1. selecting the initial of search window
Position calculates search window color probability distribution, and initial position is determined by the target that above-mentioned detection method detects, target rectangle
Circle is fixed;2. new window position is arranged with modified method, mean shift (Mean Shift) tracking is executed;3. saving zeroth order
Square;4. the size according to zeroth order away from setting search window;5. repeat step 2., 4. until convergence.
6. according to the method described in claim 1, passenger flow counting and event analysis are: 1. passenger flow counting: setting model in the picture
It encloses, will be detected when target moves into, and track the target, just stop tracking until removing setting range, count is incremented or subtracts 1;
2. people line: multiple target continues to move in detection range, and slowly;3. personnel's congestion: intensive multiple target continues one section
Time is not left in detection range;4. residue: retain original background, when motion detection does not change, and background at this time
Continue difference and decides that there is residue at scene;5. two people fight: two target Continuous move in detection range.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310459104.4A CN103559478B (en) | 2013-10-07 | 2013-10-07 | Overlook the passenger flow counting and affair analytical method in pedestrian's video monitoring |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310459104.4A CN103559478B (en) | 2013-10-07 | 2013-10-07 | Overlook the passenger flow counting and affair analytical method in pedestrian's video monitoring |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103559478A CN103559478A (en) | 2014-02-05 |
CN103559478B true CN103559478B (en) | 2018-12-04 |
Family
ID=50013719
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310459104.4A Expired - Fee Related CN103559478B (en) | 2013-10-07 | 2013-10-07 | Overlook the passenger flow counting and affair analytical method in pedestrian's video monitoring |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103559478B (en) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104091348B (en) * | 2014-05-19 | 2017-04-05 | 南京工程学院 | The multi-object tracking method of fusion marked feature and piecemeal template |
CN104008396A (en) * | 2014-05-22 | 2014-08-27 | 南京邮电大学 | In and out people flow statistical method based on people head color and shape features |
CN104537647B (en) * | 2014-12-12 | 2017-10-20 | 中安消技术有限公司 | A kind of object detection method and device |
CN106683118B (en) * | 2016-12-30 | 2020-02-14 | 北京航空航天大学 | Unmanned aerial vehicle target tracking method based on hierarchical model |
CN106845383B (en) * | 2017-01-16 | 2023-06-06 | 腾讯科技(上海)有限公司 | Human head detection method and device |
CN110008799A (en) * | 2018-11-09 | 2019-07-12 | 阿里巴巴集团控股有限公司 | Stream of people's situation evaluation method and device in specified region |
CN110120054A (en) * | 2019-05-15 | 2019-08-13 | 深圳铂石空间科技有限公司 | Automatic counting method and device, medium, electronic equipment based on image procossing |
CN110930432A (en) * | 2019-11-19 | 2020-03-27 | 北京文安智能技术股份有限公司 | Video analysis method, device and system |
CN111489474A (en) * | 2020-04-07 | 2020-08-04 | 科莱因(苏州)智能科技有限公司 | Intelligent visual tracking permission system |
CN112509338B (en) * | 2020-09-11 | 2022-02-22 | 博云视觉(北京)科技有限公司 | Method for detecting traffic jam event through silent low-point video monitoring |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1731456A (en) * | 2005-08-04 | 2006-02-08 | 浙江大学 | Bus passenger traffic statistical method based on stereoscopic vision and system therefor |
CN102521842A (en) * | 2011-11-28 | 2012-06-27 | 杭州海康威视数字技术股份有限公司 | Method and device for detecting fast movement |
CN102799894A (en) * | 2012-06-21 | 2012-11-28 | 上海理工大学 | Detection method for rapidly overlooking pedestrians |
EP2535843A1 (en) * | 2010-02-10 | 2012-12-19 | Hangzhou Hikvision Software Co. Ltd. | Method and system for population flow statistics |
CN102930287A (en) * | 2012-09-26 | 2013-02-13 | 上海理工大学 | Overlook-based detection and counting system and method for pedestrians |
-
2013
- 2013-10-07 CN CN201310459104.4A patent/CN103559478B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1731456A (en) * | 2005-08-04 | 2006-02-08 | 浙江大学 | Bus passenger traffic statistical method based on stereoscopic vision and system therefor |
EP2535843A1 (en) * | 2010-02-10 | 2012-12-19 | Hangzhou Hikvision Software Co. Ltd. | Method and system for population flow statistics |
CN102521842A (en) * | 2011-11-28 | 2012-06-27 | 杭州海康威视数字技术股份有限公司 | Method and device for detecting fast movement |
CN102799894A (en) * | 2012-06-21 | 2012-11-28 | 上海理工大学 | Detection method for rapidly overlooking pedestrians |
CN102930287A (en) * | 2012-09-26 | 2013-02-13 | 上海理工大学 | Overlook-based detection and counting system and method for pedestrians |
Non-Patent Citations (2)
Title |
---|
应用于公交客流统计的机器视觉方法;于海滨 等;《中国图形图像学报》;20080715;第13卷(第4期);第716-722页 * |
智能视频监控系统的设计及应注意的几个技术问题;雷玉堂;《中国公共安全(综合版)》;20081101;第2008年卷(第11期);第135-141页 * |
Also Published As
Publication number | Publication date |
---|---|
CN103559478A (en) | 2014-02-05 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103559478B (en) | Overlook the passenger flow counting and affair analytical method in pedestrian's video monitoring | |
CN103914688B (en) | A kind of urban road differentiating obstacle | |
CN105139425B (en) | A kind of demographic method and device | |
CN101872418B (en) | Detection method based on group environment abnormal behavior | |
US8798327B2 (en) | Method and system for people flow statistics | |
CN104303193B (en) | Target classification based on cluster | |
CN106203513B (en) | A kind of statistical method based on pedestrian's head and shoulder multi-target detection and tracking | |
CN102819764B (en) | Method for counting pedestrian flow from multiple views under complex scene of traffic junction | |
CN105427626B (en) | A kind of statistical method of traffic flow based on video analysis | |
Zhao et al. | A people counting system based on face detection and tracking in a video | |
CN104978567B (en) | Vehicle checking method based on scene classification | |
CN102982313B (en) | The method of Smoke Detection | |
CN105303191A (en) | Method and apparatus for counting pedestrians in foresight monitoring scene | |
TWI511058B (en) | A system and a method for condensing a video | |
CN105678803A (en) | Video monitoring target detection method based on W4 algorithm and frame difference | |
CN102542289A (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
CN104166861A (en) | Pedestrian detection method | |
CN106373146A (en) | Target tracking method based on fuzzy learning | |
CN108647649A (en) | The detection method of abnormal behaviour in a kind of video | |
CN106570449B (en) | A kind of flow of the people defined based on region and popularity detection method and detection system | |
CN110633678B (en) | Quick and efficient vehicle flow calculation method based on video image | |
CN105809954B (en) | Traffic incidents detection method and system | |
CN104732236B (en) | A kind of crowd's abnormal behaviour intelligent detecting method based on layered shaping | |
CN105632170A (en) | Mean shift tracking algorithm-based traffic flow detection method | |
CN103150552B (en) | A kind of driving training management method based on number of people counting |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | 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 | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20181204 Termination date: 20211007 |