CN102542289B - Pedestrian volume statistical method based on plurality of Gaussian counting models - Google Patents

Pedestrian volume statistical method based on plurality of Gaussian counting models Download PDF

Info

Publication number
CN102542289B
CN102542289B CN201110423349.2A CN201110423349A CN102542289B CN 102542289 B CN102542289 B CN 102542289B CN 201110423349 A CN201110423349 A CN 201110423349A CN 102542289 B CN102542289 B CN 102542289B
Authority
CN
China
Prior art keywords
moving target
image
target
digital
subregion
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.)
Active
Application number
CN201110423349.2A
Other languages
Chinese (zh)
Other versions
CN102542289A (en
Inventor
高陈强
余迪虎
李璐星
李强
查力
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chongqing University of Post and Telecommunications
Original Assignee
Chongqing University of Post and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Chongqing University of Post and Telecommunications filed Critical Chongqing University of Post and Telecommunications
Priority to CN201110423349.2A priority Critical patent/CN102542289B/en
Publication of CN102542289A publication Critical patent/CN102542289A/en
Application granted granted Critical
Publication of CN102542289B publication Critical patent/CN102542289B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Analysis (AREA)

Abstract

The invention relates to intelligent video surveillance and image processing and analysis, discloses a pedestrian volume statistical method, and comprises establishing a plurality of Gaussian counting models by utilizing training video sequence image samples with people number marks and performing real-time pedestrian volume statistics on videos with unknown people numbers based on the plurality of Gaussian counting models. The pedestrian volume statistical method particularly comprises the steps of firstly extracting a prospect moving target according to moving target detection, extracting eigenvectors according to moving target area and characteristics including lengths and widths of an external rectangular frame, then establishing the plurality of Gaussian counting models based on an eigenvector set, and finally analyzing numbers of pedestrians contained in an unknown moving target area based on the plurality of Gaussian counting models to achieve pedestrian volume statistics. By establishing the plurality of Gaussian counting models, the pedestrian volume statistical method avoids difficulties caused by identification and tracking of singe pedestrian, can perform statistics of the numbers of the pedestrians contained in moving target areas in different detection areas well, improves statistical accuracy of the numbers of the pedestrians, and then improves accuracy of the pedestrian volume statistics.

Description

A kind of people flow rate statistical method based on many gaussmeters digital-to-analogue type
Technical field
The present invention relates to technical field of image processing, relate in particular to a kind of method and system of people flow rate statistical.
Background technology
In existing video monitoring, great majority are just simply realized transmission of video, then rely on eye-observation to realize the work such as scene monitoring and counting.There is a large amount of weak points in this manpower monitor mode, as more uninteresting, monitor staff is easy to produce tired and causes work mistake, and in addition, along with human cost improves, relying on manpower to monitor these means of counting will be no longer applicable.
The method that people flow rate statistical system based on computer vision adopts at present can be divided three classes: the one, and the method for following the tracks of based on pedestrian detection; The 2nd, based on the method for unique point trajectory clustering; The 3rd, the method returning based on low-level feature.
The core of the method for following the tracks of based on pedestrian detection is multi-target detection, this method is to obtain foreground area by the method for background difference or machine learning, then adopts motion morphology to combine to cut apart or the method for template matches completes the task of people flow rate statistical.Under normal circumstances, this algorithm can obtain higher accuracy of detection.As the method for being mentioned in the application number Chinese patent application file that is 201010114826.2, first by sorter, present image is carried out to number of people rough detection, then rough detection result is carried out to edge feature fine screening processing, although can effectively improve the verification and measurement ratio of the number of people by said method, but, when crowd density higher, while the situation such as blocking, exist the number of people undetected, or the situation of many inspections, finally cause testing result not accurate enough, and the method calculated amount is larger, is difficult to real-time processing.
First method based on unique point trajectory clustering by following the tracks of some unique points, carries out cluster analysis and reaches the object of demographics to having the unique point track of consistent kinetic characteristic.This algorithm can effectively reduce the impact at video camera visual angle.But unique point itself is difficult to reliable and stable tracking, therefore this algorithm statistical precision is lower.
First the method returning based on low-level feature is utilized background subtraction to divide and is obtained foreground area, then calculate feature in foreground area as area, edge, texture etc., finally set up the funtcional relationship of feature and flow of the people by various regression functions as linearity, Gaussian process recurrence, neural network etc.This algorithm has been skipped the detection tracing process for single pedestrian target, has reduced computation complexity, can reach to a certain extent requirement of real-time.But its versatility is not ideal enough, and the dependence of statistical accuracy and foreground pixel extraction is larger, and therefore the method is difficult to obtain accurate number information.
In sum, in people flow rate statistical method of the prior art, difficult point is mainly how to carry out having the mobile population of greater density the statistics of degree of precision, and the complexity of algorithm can not be too high, and meets real-time application demand.
Summary of the invention
The present invention is directed to the problems referred to above that exist in the existing people flow rate statistical technology based on computer vision, a kind of real-time people flow rate statistical method based on many gaussmeters digital-to-analogue type is proposed, to solve existing people flow rate statistical scheme to the not statistical uncertainty true problem of higher density mobile population.
The present invention solves the problems of the technologies described above and adopts following technical scheme:
Based on a people flow rate statistical method for many gaussmeters digital-to-analogue type, comprising: the steps such as input picture pre-service, moving object detection, the extraction of moving target proper vector and the foundation of many gaussmeters digital-to-analogue type, motion target tracking and people flow rate statistical.Be specially:
In reality detects, often only interested in certain region in scene, therefore first choose region-of-interest (region of interest, ROI), follow-up all image processing operations all complete in this region-of-interest.Region-of-interest is divided into the detection subregion that multiple sizes are equal, adopts foreground image and background image to make the moving target detecting method of difference, obtain foreground moving target; The moving target rectangle frame that is under the jurisdiction of a connected domain is marked, extract this rectangle frame moving target proper vector and obtain set of eigenvectors, extract moving target proper vector, in same subregion, there is the target feature vector composition characteristic vector set of identical number; Set up corresponding gaussmeter digital-to-analogue type based on set of eigenvectors, the gaussmeter digital-to-analogue type composition Gauss model subset obtaining on same subregion, the model subset on all subregions forms final many gaussmeters digital-to-analogue type; When people flow rate statistical, detection line is set, the sequence of video images of unknown number is carried out to image pre-service and moving object detection; The motion target area crossing with detection line carried out to target following, judge that whether moving target boundary rectangle frame is crossing with detection line, if non-intersect, next frame image is processed, until its boundary rectangle frame arrives detection line, if intersect, extract the proper vector of current moving target; In each two field picture of tracing process, extract the proper vector of current moving target, according to the residing subregion of current moving target, adopt corresponding Gauss model subset to analyze the number in current moving target, adopt fast target to follow the tracks of correlating method and obtain count queue corresponding to this target area, and deposit the number obtaining the corresponding queue of in this target area; In the time that moving target leaves detection line, calculate the mean value of number in queue, obtain people flow rate statistical.
Region-of-interest in scene is divided into a series of detection subregions, as may be partitioned into
Figure 190410DEST_PATH_IMAGE001
the detection subregion that individual area is identical, adopts
Figure 175684DEST_PATH_IMAGE002
represent the
Figure 672393DEST_PATH_IMAGE003
oK,
Figure 743117DEST_PATH_IMAGE004
row subregion (
Figure 734207DEST_PATH_IMAGE005
,
Figure 573987DEST_PATH_IMAGE006
), wherein
Figure 54647DEST_PATH_IMAGE007
,
Figure 47222DEST_PATH_IMAGE008
according to the size of field of detection, and pedestrian's size in visual field, confirm its value.Generally, detect subregion size and can be three to four pedestrian's sizes in field of detection.Video image after cutting apart is carried out to the disposal of gentle filter, reduce the impact of noise.
Extract background image, current frame image and background image in detection subregion are carried out to difference processing, obtain difference image.
The standard variance of pixel value between different gray areas in calculating difference image, and determine segmentation threshold according to the maximal value in these standard variances, to Image Segmentation Using and the processing of bianry image morphology, extract foreground moving target.
Its boundary rectangle frame of moving target that is under the jurisdiction of a connected domain is marked, and extracts the proper vector of this motion target area
Figure 904320DEST_PATH_IMAGE009
.
According to the residing position of rectangle frame central point, determine moving target the inferior detection subregion that occurs place is
Figure 187851DEST_PATH_IMAGE002
, pedestrian's number is
Figure 420118DEST_PATH_IMAGE011
time proper vector
Figure 815327DEST_PATH_IMAGE012
(in the embodiment of the present invention, comprising moving target area is
Figure 301803DEST_PATH_IMAGE013
(number of pixels), rectangle frame is long
Figure 389845DEST_PATH_IMAGE014
, wide three features), will there is the proper vector composition characteristic vector set of identical pedestrian's number
Figure 278615DEST_PATH_IMAGE016
.Set up corresponding gaussmeter digital-to-analogue type based on set of eigenvectors
Figure 885176DEST_PATH_IMAGE017
.Be specially:
Adopt formula:
Figure 144119DEST_PATH_IMAGE018
(wherein representing to detect subregion is
Figure 822411DEST_PATH_IMAGE002
, detection number is
Figure 345797DEST_PATH_IMAGE011
time, corresponding sampling feature vectors number), calculated characteristics vector set
Figure 447745DEST_PATH_IMAGE016
mean vector , adopt formula according to mean vector:
Figure 918489DEST_PATH_IMAGE020
calculate its covariance matrix
Figure 296380DEST_PATH_IMAGE021
, according to formula:
Figure 834809DEST_PATH_IMAGE022
, foundation detects subregion and is
Figure 298151DEST_PATH_IMAGE002
, detection pedestrian number is
Figure 829496DEST_PATH_IMAGE011
time, corresponding gaussmeter digital-to-analogue type is
Figure 327473DEST_PATH_IMAGE017
.
By the gaussmeter digital-to-analogue type of corresponding same detection subregion
Figure 771224DEST_PATH_IMAGE017
composition gaussmeter digital-to-analogue type subset
Figure 987442DEST_PATH_IMAGE023
, by multiple gaussmeter digital-to-analogue type subsets
Figure 869947DEST_PATH_IMAGE023
form many gaussmeters digital-to-analogue type
Figure 910846DEST_PATH_IMAGE024
.
While carrying out people flow rate statistical, detection line is set, the sequence of video images of unknown number is carried out to image pre-service and moving object detection, judge that whether moving target boundary rectangle frame is crossing with detection line, if non-intersect, next frame image is processed, until its boundary rectangle frame arrives detection line, if intersect, extract the proper vector of current moving target, according to its residing subregion
Figure 587815DEST_PATH_IMAGE002
, call corresponding gaussmeter digital-to-analogue type subset
Figure 963433DEST_PATH_IMAGE023
calculate the pedestrian's number comprising in current moving target, and deposit the number obtaining in queue that this moving target is corresponding; Continue pursuit movement target, in subsequent frame, call corresponding gaussmeter digital-to-analogue type subset according to its residing subregion, calculate pedestrian's number that this moving target comprises and deposit corresponding queue in; When moving target leaves detection line, carry out people flow rate statistical, that is, calculate the average of number in queue, obtain the number of current tracking target.
The present invention, by the moving object detection in region-of-interest, sets up gaussmeter digital-to-analogue type and carries out the calculating of moving target number, can in the time of the flow of the people of large density, carry out the people flow rate statistical of degree of precision, and reduce computation complexity, reaches requirement of real-time.
Accompanying drawing explanation
The process flow diagram of the people flow rate statistical method based on many gaussmeters digital-to-analogue type in Fig. 1 embodiment of the present invention;
Many gaussmeters number modelling phase process flow diagram in Fig. 2 embodiment of the present invention;
Foreground moving target detection process flow diagram in Fig. 3 embodiment of the present invention;
Target following schematic diagram in Fig. 4 embodiment of the present invention.
Embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in detail.
A kind of people flow rate statistical based on many gaussmeters digital-to-analogue type that the present invention proposes is for real-time monitoring system.Obtain foreground moving target by motion detection, according to the number in many gaussmeters digital-to-analogue type analysis present image, realize people flow rate statistical.
Fig. 1 is the people flow rate statistical process flow diagram based on many gaussmeters digital-to-analogue type in the embodiment of the present invention.As shown in Figure 1, in the embodiment of the present invention carrying out needing to set up many gaussmeters digital-to-analogue type before people flow rate statistical.
Region-of-interest in scene is divided into a series of detection subregions.Video image after cutting apart is carried out to the disposal of gentle filter.
Extract background image, current frame image and background image in detection subregion are carried out to difference processing, obtain difference image.
The standard variance of pixel value between different gray areas in calculating difference image, and determine segmentation threshold according to the maximal value in these standard variances, to Image Segmentation Using and the processing of bianry image morphology, extract foreground moving target.
Its boundary rectangle frame of moving target that is under the jurisdiction of a connected domain is marked, and extracts the proper vector of this motion target area
Figure 649629DEST_PATH_IMAGE009
.
According to the residing position of rectangle frame central point, determine moving target
Figure 43570DEST_PATH_IMAGE010
the inferior detection subregion that occurs place is
Figure 157020DEST_PATH_IMAGE002
, pedestrian's number is
Figure 19934DEST_PATH_IMAGE011
time proper vector
Figure 244242DEST_PATH_IMAGE012
(in the embodiment of the present invention, proper vector comprises moving target area and is (number of pixels), rectangle frame is long , wide
Figure 628715DEST_PATH_IMAGE015
three features), will there is the proper vector composition characteristic vector set of identical pedestrian's number
Figure 656714DEST_PATH_IMAGE016
, set up corresponding gaussmeter digital-to-analogue type based on set of eigenvectors
Figure 25247DEST_PATH_IMAGE017
.Be specially:
Adopt formula:
Figure 418182DEST_PATH_IMAGE018
(wherein
Figure 318005DEST_PATH_IMAGE008
representing to detect subregion is
Figure 572531DEST_PATH_IMAGE002
, detection number is
Figure 608620DEST_PATH_IMAGE011
time, corresponding sampling feature vectors number), calculated characteristics vector set
Figure 172457DEST_PATH_IMAGE016
mean vector
Figure 293997DEST_PATH_IMAGE019
, adopt formula according to mean vector:
Figure 850749DEST_PATH_IMAGE020
calculate its covariance matrix
Figure 6924DEST_PATH_IMAGE021
, according to formula:
Figure 741661DEST_PATH_IMAGE022
, foundation detects subregion and is
Figure 350497DEST_PATH_IMAGE002
, detection pedestrian number is
Figure 946826DEST_PATH_IMAGE011
time, corresponding gaussmeter digital-to-analogue type is
Figure 895190DEST_PATH_IMAGE017
.
By the gaussmeter digital-to-analogue type of corresponding same detection subregion
Figure 863146DEST_PATH_IMAGE017
composition gaussmeter digital-to-analogue type subset , by multiple gaussmeter digital-to-analogue type subsets form many gaussmeters digital-to-analogue type .
While carrying out people flow rate statistical, detection line is set, the sequence of video images of unknown number is carried out to image pre-service and moving object detection, judge that whether moving target boundary rectangle frame is crossing with detection line, if non-intersect, next frame image is processed, until its boundary rectangle frame arrives detection line, if intersect, extract the proper vector of current moving target, according to its residing subregion
Figure 65141DEST_PATH_IMAGE002
, call corresponding gaussmeter digital-to-analogue type subset
Figure 71405DEST_PATH_IMAGE023
calculate the pedestrian's number comprising in current moving target, and deposit the number obtaining in queue that this moving target is corresponding; Continue pursuit movement target, in subsequent frame, call corresponding gaussmeter digital-to-analogue type subset according to its residing subregion, calculate pedestrian's number that this moving target comprises and deposit corresponding queue in; When moving target leaves detection line, carry out people flow rate statistical, that is, calculate the average of number in queue, obtain the number of current tracking target.
Fig. 2 is many gaussmeters digital-to-analogue type Establishing process figure in the embodiment of the present invention, specifically comprises following several step:
Obtain the sequence of video images with number mark by shooting.
Surveyed area is divided.The region-of-interest of selecting the video image of input, is then divided into region-of-interest a series of detection subregions, region-of-interest in scene is divided into a series of detection subregions, as may be partitioned into
Figure 586700DEST_PATH_IMAGE001
the detection subregion that individual area is identical, adopts
Figure 509657DEST_PATH_IMAGE002
represent the
Figure 819415DEST_PATH_IMAGE025
oK,
Figure 811511DEST_PATH_IMAGE004
row subregion (
Figure 864918DEST_PATH_IMAGE005
,
Figure 704698DEST_PATH_IMAGE006
), wherein
Figure 123041DEST_PATH_IMAGE007
,
Figure 681061DEST_PATH_IMAGE008
according to the size of field of detection, and pedestrian's size in visual field, confirm its value.Generally, detect subregion size and can be three to four pedestrian's sizes in field of detection.Video image after cutting apart is carried out to the disposal of gentle filter, reduce the impact of noise.
Image pre-service.
The input video sequence image of region-of-interest is converted into grayscale image sequence, and pretreatment module is carried out filtering to grayscale image sequence, the noise in removal of images.The embodiment of the present invention adopts Gaussian smoothing to carry out filtering to image.
Moving object detection.Extract background image: according to the feature of monitoring scene, select suitable background image to extract.Current frame image and background image are carried out to calculus of differences acquisition difference image, carry out carrying out image threshold segmentation and the processing of bianry image morphology, obtain foreground moving target.
Proper vector is extracted.
Obtain moving target area in rectangle frame, and the length of rectangle frame and width.Take moving target area as (
Figure 949276DEST_PATH_IMAGE013
), rectangle frame length (
Figure 909142DEST_PATH_IMAGE014
), wide (
Figure 498386DEST_PATH_IMAGE015
) three latent structure proper vectors,
Figure 278123DEST_PATH_IMAGE026
. represent the
Figure 674655DEST_PATH_IMAGE010
the subregion at the moving target place of inferior appearance is , pedestrian's number is
Figure 701834DEST_PATH_IMAGE011
time proper vector;
The target connected domain that in same subregion, number is identical (is had identical ) proper vector
Figure 258029DEST_PATH_IMAGE030
the set of composition characteristic vector .
Analyze each proper vector and concentrate vectorial number, as vectorial number is less than
Figure 943405DEST_PATH_IMAGE008
, continue to extract proper vector, wherein
Figure 945996DEST_PATH_IMAGE008
value be the bigger the better, rule of thumb generally get
Figure 469381DEST_PATH_IMAGE031
.Concentrating vectorial number when proper vector is more than or equal to
Figure 820597DEST_PATH_IMAGE008
, according to set of eigenvectors, set up single gaussmeter digital-to-analogue type
Figure 62223DEST_PATH_IMAGE017
.Single gaussmeter digital-to-analogue type process of establishing as follows:
Call formula:
Figure 918500DEST_PATH_IMAGE018
calculated characteristics vector set
Figure 207661DEST_PATH_IMAGE016
mean vector
Figure 671004DEST_PATH_IMAGE019
, utilize formula according to mean vector:
Figure 953080DEST_PATH_IMAGE020
calculate its covariance matrix
Figure 185479DEST_PATH_IMAGE021
, set up surveyed area according to covariance matrix and be
Figure 144076DEST_PATH_IMAGE002
, detection pedestrian number is
Figure 94715DEST_PATH_IMAGE011
time, corresponding gaussmeter digital-to-analogue type :
Figure 267387DEST_PATH_IMAGE022
, in formula,
Figure 209935DEST_PATH_IMAGE032
for random vector, represent transposition.
When the vector set to all set up after corresponding gaussmeter digital-to-analogue type, (have identical in same subregion
Figure 167155DEST_PATH_IMAGE034
) gaussmeter digital-to-analogue type
Figure 467555DEST_PATH_IMAGE017
composition gaussmeter digital-to-analogue type subset
Figure 392786DEST_PATH_IMAGE023
, all gaussmeter digital-to-analogue type subsets
Figure 554777DEST_PATH_IMAGE023
form many gaussmeters digital-to-analogue type ,
Figure 900625DEST_PATH_IMAGE035
.
Complete after the foundation of many gaussmeters digital-to-analogue type, can carry out people flow rate statistical according to this counting model.As shown in Figure 1, specifically comprise the steps:
Obtain video sequence image by single ccd imaging sensor by vertical shooting.Surveyed area is carried out to region division.
Detection line is set, preferably selects the center line of region-of-interest, as the line segment L1 of Fig. 4.Image is carried out to pre-service, moving object detection.
If moving target do not detected, go to next frame and continue to process; If moving target detected, judge that whether moving target boundary rectangle frame is crossing with detection line, if non-intersect, go to next frame and continue processing, otherwise extract the proper vector of current moving target
Figure 1567DEST_PATH_IMAGE036
(the same with the proper vector of many gaussmeters number modelling phase).Estimate the number that current motion target area comprises, specifically comprise the steps:
Determine its residing subregion according to moving target boundary rectangle frame center point coordinate
Figure 29566DEST_PATH_IMAGE002
, in many gaussmeters digital-to-analogue type, obtain subregion
Figure 148832DEST_PATH_IMAGE002
corresponding gaussmeter digital-to-analogue type subset
Figure 604084DEST_PATH_IMAGE023
,
Figure 425278DEST_PATH_IMAGE023
for
Figure 991389DEST_PATH_IMAGE037
, ,
Figure 856894DEST_PATH_IMAGE039
,
Figure 978433DEST_PATH_IMAGE040
the set of composition.Current proper vector
Figure 36650DEST_PATH_IMAGE036
bring respectively model (Gaussian function) into
Figure 192825DEST_PATH_IMAGE037
,
Figure 927563DEST_PATH_IMAGE038
,
Figure 536399DEST_PATH_IMAGE039
,
Figure 631263DEST_PATH_IMAGE040
middle calculating, result of calculation is peaked model
Figure 641944DEST_PATH_IMAGE017
corresponding number is the number that current motion target area comprises, and its value equals
Figure 609900DEST_PATH_IMAGE011
.That is:
Figure 355319DEST_PATH_IMAGE042
Calculate the number comprising in current moving target
Figure 908923DEST_PATH_IMAGE043
.Wherein,
Figure 47780DEST_PATH_IMAGE044
represent model subset
Figure 568891DEST_PATH_IMAGE045
in maximal value.
After the number that acquisition motion target area comprises, can pass through the fast tracking method based on detection line, obtain count queue corresponding to this target area, and current acquisition Population size estimation is deposited in this count queue.Statistical number of person is recorded in queue corresponding to this detection target.
Be illustrated in figure 3 foreground moving target detection process flow diagram, specifically comprise the steps:
Extract background image: according to the feature of monitoring scene, select suitable background image extracting method.As: do not change or change very little particular surroundings (as part indoor environment) for some backgrounds, directly shooting background image, then keeps background image constant.For the obvious scene of change of background (as natural scene), use and obtain background image based on histogrammic background modeling method, be specially: the grey level histogram of statistical series image pixel value on same pixel position, the maximum gray-scale value of occurrence number is as the background pixel value of this point.
Obtain difference image: in moving object detection processing module, the current image frame of obtaining and background image are carried out to calculus of differences from sequence of video images, obtain difference image.
Difference image Threshold segmentation: 8 gray level images that the difference image obtaining is standard, therefore its pixel value scope is [0,255], ask for successively tonal range in difference image and be respectively [0,255], [1,255] ..., [254,255], the standard variance of pixel values in 254 gray areas altogether, then calculates the maximal value of these 254 standard variances
Figure 84186DEST_PATH_IMAGE046
, and two threshold values are set
Figure 803880DEST_PATH_IMAGE047
,
Figure 300590DEST_PATH_IMAGE048
(
Figure 371314DEST_PATH_IMAGE049
), (
Figure 362404DEST_PATH_IMAGE047
,
Figure 202184DEST_PATH_IMAGE048
can choose according to inputted video image quality, generally, ,
Figure 929279DEST_PATH_IMAGE051
), according to
Figure 520798DEST_PATH_IMAGE047
,
Figure 152767DEST_PATH_IMAGE048
in the middle of determining, threshold value T, asks for final segmentation threshold by following formula ;
Figure 36596DEST_PATH_IMAGE053
，
Figure 431805DEST_PATH_IMAGE054
，
Wherein
Figure 183860DEST_PATH_IMAGE044
, represent to ask for respectively maximal value and the minimum value of two numerical value.With
Figure 273356DEST_PATH_IMAGE052
as threshold value, to Image Segmentation Using, obtain bianry image.
Bianry image morphology processing: carry out morphologic filtering in bianry image, operate and delete the less false target region of some areas by morphological erosion, and by expansive working, region merging is carried out in the target area of some fractures.Then, adopt 8 each connected regions of neighbour's connective region search algorithm search, carry out mark with the boundary rectangle of connected domain, obtain thus foreground moving target.
Target fast tracking method as shown in Figure 4, specifically comprise the steps: that in the present embodiment, the object of target following is exactly to realize the association of motion target area in two continuous frames image, before and after judging, in two two field pictures, whether the moving target crossing with detection line is same target.
In Fig. 4, show under two nearer target conditions of single moving target and longitudinal separation the state in two two field pictures of front and back.In figure, region R1 is video-input image; Region R2 is region-of-interest (can select according to actual needs the region-of-interest of suitable shape and size); Region R3 is the boundary rectangle frame of moving target; Line segment L1 is detection line, generally selects the center line of region-of-interest; Line segment L2 is the intersection section of moving target and detection line; P1, P2 is front and back two frames (frames and
Figure 172390DEST_PATH_IMAGE056
frame) summit in the upper right corner of external rectangle frame in image.
The coordinate of supposing P1 and P2 is respectively
Figure 841269DEST_PATH_IMAGE057
with
Figure 772316DEST_PATH_IMAGE058
(image coordinate system, be that the image upper left corner is initial point, X-axis positive dirction be level to the right, Y-axis positive dirction is for vertically downward), whether the moving target before and after differentiating by two constraint conditions in two two field pictures is same target, that is: before and after (1), the intersection section of the motion target area in two two field pictures and detection line is overlapping or partly overlap; (2) coordinate .If the moving target in two two field pictures of front and back meets (1) and (2) two constraint conditions simultaneously, be judged as same target, no person is judged as different targets.For situation about only comprising when a moving target, can realize correct judgement with retraining (1), as shown in Figure 4 (a).But to the situation of two nearer moving targets of the longitudinal separation of Fig. 4 (b), although the intersection section of front and back two frames is overlapping, must could realize correct judgement by constraint (2).Owing to processing, frame frequency is higher, and (1) and (2) two constraint Rule of judgment can be realized large absolutely correct judgement of counting situation, and this algorithm is simple, efficient.
If moving target leaves detection line, continuous three frames of moving target and detection line when non-intersect, meet counting condition, calculate the average of pedestrian's number in the count queue that current moving target is corresponding
Figure 450608DEST_PATH_IMAGE060
, obtain the number that current moving target comprises.Then the average of cumulative each queue pedestrian number, obtains total flow of the people
Figure 973993DEST_PATH_IMAGE061
thereby, realize people flow rate statistical.
The above is only the preferred embodiment of the present invention; be noted that for those skilled in the art, under the premise without departing from the principles of the invention; can also make some improvements and modifications, these improvements and modifications also should be considered as protection scope of the present invention.

Claims (3)

1. the people flow rate statistical method based on many gaussmeters digital-to-analogue type, it is characterized in that, comprise the steps, be detection subregion by Video Image Segmentation, current frame image and background image are done to difference, obtain difference image, extract background image, calculate in the difference image that detects present image and background image in subregion the maximal value σ of the standard variance of pixel value between each gray area max, and according to formula:
T=max (T 1, σ max), Th s=min (T 2, T) and ask for segmentation threshold Th s, with Th sas threshold value, to Image Segmentation Using, obtain bianry image, bianry image morphology is processed and is obtained foreground moving target, wherein, T 1, T 2for the threshold value arranging, meet and be related to T 2>T 1>0; The moving target rectangle frame that is under the jurisdiction of a connected domain is marked, extracts this rectangle frame moving target proper vector and obtain set of eigenvectors; Appear at for the n time and detect subregion as R take moving target area s, the long h of rectangle frame, tri-latent structure moving targets of wide w i,j, proper vector when pedestrian's number is k to there is the proper vector composition characteristic vector set of identical pedestrian's number, call formula according to proper vector:
Figure FDA0000408341930000012
the mean vector μ of calculated characteristics vector set i, j, k, utilize formula according to mean vector:
Figure FDA0000408341930000013
calculate covariance matrix C i, j, k, according to formula: G i , j , k ( x ) = 1 ( 2 π ) 3 / 2 | C i , j , k | 1 / 2 exp { - 1 2 ( x - μ i , j , k ) T C i , j , k - 1 ( x - μ i , j , k ) } Set up surveyed area for (i, j), when detection pedestrian number is k, corresponding gaussmeter digital-to-analogue type is G i, j, k, and the G in same subregion i, j, kbe classified as a gaussmeter digital-to-analogue type subset G i,j, all gaussmeter digital-to-analogue type subsets form many gaussmeters digital-to-analogue type, and in formula, x is random vector, and T represents transposition, and N is sampling feature vectors number; Judge that whether moving target boundary rectangle frame is crossing with detection line, if non-intersect, next frame image is processed, until its boundary rectangle frame arrives detection line; If intersect, extract the proper vector of current moving target, according to the residing subregion of current moving target, determine residing subregion R according to moving target boundary rectangle frame center point coordinate i,j, in many gaussmeters digital-to-analogue type, obtain subregion R i,jcorresponding gaussmeter digital-to-analogue type subset G i,j, utilize formula: NUM=k if ( G i , j , k ( F ) = max ( G i , j , 1 ( F ) , G i , j , 2 ( F ) , · · · , G i , j , N p ( F ) ) ) Calculate the number comprising in current moving target.
2. people flow rate statistical method according to claim 1, it is characterized in that, after the number that acquisition motion target area comprises, adopt fast target to follow the tracks of correlating method and obtain count queue corresponding to this target area, and deposit current moving target number in this queue, in the time that target is left detection line, the mean value that calculates queue obtains the number that this target area comprises.
3. people flow rate statistical method according to claim 2, is characterized in that, fast target is followed the tracks of correlating method and is specially: suppose that P1 and P2 are respectively the summit, the upper right corner of moving target boundary rectangle in two two field pictures of front and back, its coordinate is respectively (x 1, y 1) and (x 2, y 2), as satisfy condition: the motion target area before and after (1) in two two field pictures and the intersection section of detection line are overlapping or partly overlap; (2) coordinate x 1<x 2, be same target.
CN201110423349.2A 2011-12-16 2011-12-16 Pedestrian volume statistical method based on plurality of Gaussian counting models Active CN102542289B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201110423349.2A CN102542289B (en) 2011-12-16 2011-12-16 Pedestrian volume statistical method based on plurality of Gaussian counting models

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201110423349.2A CN102542289B (en) 2011-12-16 2011-12-16 Pedestrian volume statistical method based on plurality of Gaussian counting models

Publications (2)

Publication Number Publication Date
CN102542289A CN102542289A (en) 2012-07-04
CN102542289B true CN102542289B (en) 2014-06-04

Family

ID=46349139

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201110423349.2A Active CN102542289B (en) 2011-12-16 2011-12-16 Pedestrian volume statistical method based on plurality of Gaussian counting models

Country Status (1)

Country Link
CN (1) CN102542289B (en)

Families Citing this family (37)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103839035A (en) * 2012-11-22 2014-06-04 富士通株式会社 Person number statistical method and person number statistical system
CN103186903A (en) * 2013-04-22 2013-07-03 苏州科技学院 Moving object detecting and tracking system based on PID (Proportion Integration Differentiation)
CN104239905A (en) * 2013-06-17 2014-12-24 上海盖普电梯有限公司 Moving target recognition method and intelligent elevator billing system having moving target recognition function
CN103400113B (en) * 2013-07-10 2016-08-24 重庆大学 Freeway tunnel pedestrian detection method based on image procossing
CN104318240B (en) * 2014-07-18 2017-11-07 天津理工大学 A kind of bud method of discrimination based on computer vision
CN104123569B (en) * 2014-07-21 2017-02-15 中国人民解放军63908部队 Video person number information statistics method based on supervised learning
CN104506820A (en) * 2015-01-06 2015-04-08 成都新舟锐视科技有限公司 Intelligent multi-target tracing system
EP3059722A1 (en) * 2015-02-20 2016-08-24 Airbus Group India Private Limited Management of aircraft in-cabin activities occuring during turnaround using video analytics
CN104965964B (en) * 2015-08-06 2018-01-30 山东建筑大学 A kind of building personnel's distributed model method for building up based on monitor video analysis
CN105551266B (en) * 2015-12-08 2017-09-29 合肥寰景信息技术有限公司 A kind of computational methods of the flow of the people threshold value of control traffic signaling equipment
CN105574499B (en) * 2015-12-15 2019-08-20 东华大学 A kind of number detection statistics method and system based on SOC
CN107180420A (en) * 2016-03-09 2017-09-19 顺丰科技有限公司 Article transmits householder method
CN106407946B (en) * 2016-09-29 2020-03-03 北京市商汤科技开发有限公司 Cross-line counting method, deep neural network training method, device and electronic equipment
CN106600576B (en) * 2016-10-25 2020-02-18 深圳秀蛋科技有限公司 Human head locking method based on RGB-D camera
CN106600643B (en) * 2016-10-25 2019-06-28 长安大学 A kind of demographic method based on trajectory analysis
CN106600590B (en) * 2016-12-12 2019-09-06 四川华雁信息产业股份有限公司 Current transformer oil leak automatic testing method and device
CN106845620B (en) * 2016-12-19 2019-09-10 江苏慧眼数据科技股份有限公司 A kind of passenger flow counting method based on quene state analysis
CN106846297A (en) * 2016-12-21 2017-06-13 深圳市镭神智能系统有限公司 Pedestrian's flow quantity detecting system and method based on laser radar
CN106845374B (en) * 2017-01-06 2020-03-27 清华大学 Pedestrian detection method and detection device based on deep learning
CN109643321A (en) * 2017-02-15 2019-04-16 富士通株式会社 Video analytic system and video analysis method based on video monitoring
CN107123126A (en) * 2017-03-29 2017-09-01 天棣网络科技(上海)有限公司 A kind of stream of people's moving scene temperature method of estimation
CN109389016B (en) * 2017-08-10 2023-04-07 株式会社日立制作所 Method and system for counting human heads
CN107729804A (en) * 2017-08-31 2018-02-23 广东数相智能科技有限公司 A kind of people flow rate statistical method and device based on garment ornament
CN109753842B (en) * 2017-11-01 2021-07-16 深圳先进技术研究院 People flow counting method and device
CN107770921A (en) * 2017-12-07 2018-03-06 四川谋席科技有限公司 A kind of energy-saving LED illumination system based on Cloud Server
CN107809834A (en) * 2017-12-07 2018-03-16 四川谋席科技有限公司 A kind of resident illuminator based on cloud computing
CN108229368B (en) * 2017-12-28 2020-05-26 浙江大华技术股份有限公司 Video display method and device
CN108549835A (en) * 2018-03-08 2018-09-18 深圳市深网视界科技有限公司 Crowd counts and its method, terminal device and the storage medium of model construction
CN110826581B (en) * 2018-08-10 2023-11-07 京东科技控股股份有限公司 Animal number identification method, device, medium and electronic equipment
CN109800683A (en) * 2018-12-30 2019-05-24 昆明物理研究所 A kind of infrared pedestrian detection method and device based on FPGA
CN110008766B (en) * 2019-03-26 2020-04-24 南京大学 Indoor people counting method based on radio frequency identification technology
CN111260628A (en) * 2020-01-15 2020-06-09 北京林业大学 Large nursery stock number counting method based on video image and electronic equipment
CN111680569B (en) * 2020-05-13 2024-04-19 北京中广上洋科技股份有限公司 Attendance rate detection method, device, equipment and storage medium based on image analysis
CN112005247A (en) * 2020-06-12 2020-11-27 深圳盈天下视觉科技有限公司 People flow data monitoring system and people flow data display method and device thereof
CN111739024B (en) * 2020-08-28 2020-11-24 安翰科技(武汉)股份有限公司 Image recognition method, electronic device and readable storage medium
CN113536891B (en) * 2021-05-10 2023-01-03 新疆爱华盈通信息技术有限公司 Pedestrian traffic statistical method, storage medium and electronic equipment
CN114943930A (en) * 2022-05-05 2022-08-26 北京埃福瑞科技有限公司 Station passenger clearing method and device

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6674877B1 (en) * 2000-02-03 2004-01-06 Microsoft Corporation System and method for visually tracking occluded objects in real time
CN101325690A (en) * 2007-06-12 2008-12-17 上海正电科技发展有限公司 Method and system for detecting human flow analysis and crowd accumulation process of monitoring video flow
CN101303727B (en) * 2008-07-08 2011-11-23 北京中星微电子有限公司 Intelligent management method based on video human number Stat. and system thereof

Also Published As

Publication number Publication date
CN102542289A (en) 2012-07-04

Similar Documents

Publication Publication Date Title
CN102542289B (en) Pedestrian volume statistical method based on plurality of Gaussian counting models
CN107527009B (en) Remnant detection method based on YOLO target detection
CN108986064B (en) People flow statistical method, equipment and system
CN101389004B (en) Moving target classification method based on on-line study
CN106447680B (en) The object detecting and tracking method that radar is merged with vision under dynamic background environment
CN103824070B (en) A kind of rapid pedestrian detection method based on computer vision
CN103164858B (en) Adhesion crowd based on super-pixel and graph model is split and tracking
CN105404847B (en) A kind of residue real-time detection method
CN102214309B (en) Special human body recognition method based on head and shoulder model
CN104978567B (en) Vehicle checking method based on scene classification
CN104166841A (en) Rapid detection identification method for specified pedestrian or vehicle in video monitoring network
CN103310194A (en) Method for detecting head and shoulders of pedestrian in video based on overhead pixel gradient direction
CN102521565A (en) Garment identification method and system for low-resolution video
CN105512618B (en) Video tracing method
CN104992453A (en) Target tracking method under complicated background based on extreme learning machine
CN110991397B (en) Travel direction determining method and related equipment
CN102768726A (en) Pedestrian detection method for preventing pedestrian collision
CN103871081A (en) Method for tracking self-adaptive robust on-line target
CN117994987B (en) Traffic parameter extraction method and related device based on target detection technology
CN109684986B (en) Vehicle analysis method and system based on vehicle detection and tracking
CN104599291B (en) Infrared motion target detection method based on structural similarity and significance analysis
CN106529441B (en) Depth motion figure Human bodys&#39; response method based on smeared out boundary fragment
Su et al. A new local-main-gradient-orientation HOG and contour differences based algorithm for object classification
CN103049788A (en) Computer-vision-based system and method for detecting number of pedestrians waiting to cross crosswalk
CN106056078A (en) Crowd density estimation method based on multi-feature regression ensemble learning

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant