CN107153819A - A kind of queue length automatic testing method and queue length control method - Google Patents
A kind of queue length automatic testing method and queue length control method Download PDFInfo
- Publication number
- CN107153819A CN107153819A CN201710312890.3A CN201710312890A CN107153819A CN 107153819 A CN107153819 A CN 107153819A CN 201710312890 A CN201710312890 A CN 201710312890A CN 107153819 A CN107153819 A CN 107153819A
- Authority
- CN
- China
- Prior art keywords
- mrow
- queue
- queue length
- pedestrian
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of queue length automatic testing method and queue length control method, and pedestrian is detected based on organization of human body model algorithm；Using feature on-line selection boosting algorithm tracking pedestrians；According to the people detected, the queue shape of people is fitted using the least square polynomial curve fitting method based on RANSAC, so as to estimate the length of queue；Realize that image coordinate is tied to the conversion of world coordinate system by image calibration, calculate the actual length queued up；According to curve matching calculated curve length, the distance between phase neighbour in the length and queue further according to queuing calculates the number in queue.The queue length automatic testing method and queue length control method of the present invention can fast and accurately detect pedestrian's queue length in monitoring scene, it is to avoid the problem of target is lost, and reduce missing inspection；By monitoring queuing situation in real time, window is handled in adjustment in time, so that the problem of effectively solving to wait in line, preferably serves the passengers.
Description
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of queue length automatic testing method and length of queuing up
Spend control method.
Background technology
In the PEs such as airport, passenger is when checking in board and entering security check, it is often necessary to wait in line, by
Compare concentration in personnel, passenger's queue time will extend when number is more, so as to cause the impatience of passenger；Together
When, passenger needs to arrive at the airport for a long time with the airliner that avoids delay in advance, and greatly reduce passenger seizes the opportunity experience.In hospital
Problem above is equally existed Deng public arena.
Existing solution be the staff on airport by manually dredging, reduce the queue time of passenger；By
In manually dredging completely by staff to the control of field condition, waste time and energy, and efficiency is also at a fairly low.
Therefore, how preferably to serve the passengers, reduce passenger's queue time, improve airport operating efficiency and reduce manpower
Resource overhead turns into one of those skilled in the art's urgent problem to be solved.
The content of the invention
The shortcoming of prior art in view of the above, it is an object of the invention to provide a kind of queue length automatic detection side
Method and queue length control method, for solving manually to dredge the problem of wasting time and energy in the prior art.
In order to achieve the above objects and other related objects, the present invention provides a kind of queue length automatic testing method, described
Queue length automatic testing method at least includes：
Step S1：Obtain pedestrian's queuing video, pedestrian's queuing video described in typing；
Step S2：Pedestrian in pedestrian's queuing video is detected based on organization of human body model algorithm；
Step S3：Detect after pedestrian, feature based on-line selection boosting algorithm constantly updates pedestrian image characteristic value, with
Positioning is tracked to pedestrian；
Step S4：According to the pedestrian detected, the queue pattern curve of pedestrian is fitted；
Step S5：Image calibration, world coordinates is converted to by image coordinate；
Step S6：The distance between physical length of queue and adjacent pedestrian are calculated according to the queue pattern curve, from
And calculate the number obtained in queue.
Preferably, step S2 is specifically included：One two dimensional model based on pedestrian's part is constructed according to organization of human body, by carrying
The architectural feature of pedestrian image is taken to be matched with two dimensional model, so as to recognize pedestrian.
Preferably, the online boosting algorithm further comprises：N number of feature selector, is used as strong classifier；M feature
A feature pool is constituted, Weak Classifier is used as；When new samples are reached, N number of feature selector is sequentially generated, and generation is to M every time
The correct sample weights of accumulation classification and accumulation classification error sample weights of individual feature are updated, and each feature selector will be currently
The feature of minimum cumulative error frequency combines to form strong classifier as its corresponding Weak Classifier, N number of feature selector, and target is new
Position in the environs of target location in previous frame with strong classifier evaluation by being determined.
It is highly preferred that the correct sample weights of accumulation classification and accumulation classification error sample weights to M feature are carried out more
New method is specifically included：
Work as hm(x)=y when,
Work as hm(x) ≠ y when,
Wherein, hm, (m=1 ..., M) is characterized, and x is the video in window with the size such as target area, and y is used to represent just
Sample or negative sample, y=0 are negative sample, and y=1 is positive sample,To accumulate the correct sample weights of classification,For accumulation
Classification error sample weights, λ is the current weights of sample, and initial value is 1.
Formed it is highly preferred that the strong classifier is combined by N number of feature selector by weight：
Weight αnMeet following relation：
Wherein,It is characterized selector, εnIt is characterized the corresponding cumulative error frequencies of n.
It is highly preferred that target new position meets following relation：
Wherein,pw(w=1 ..., W) is the target candidate position in hunting zone,
It is characterized selector, αnFor weight.
It is highly preferred that the feature in the online boosting algorithm includes：Haar features, edge orientation histogram feature and base
In the feature of the local binary patterns of the block information in space.
It is highly preferred that the computational methods of the Haar features use integration histogram.
It is highly preferred that the computational methods of the edge orientation histogram feature are as follows：By the direction θ of pixel gradient in area
Between θ ∈ (- π, π] in be quantified as multiple angular regions, and the gradient of each pixel in region is counted, by corresponding width
Value adds corresponding angle and obtains histogram.
It is highly preferred that the direction of the pixel gradient meets following relation：
Wherein, gradient magnitude is extracted in the horizontal directionThe amplitude of gradient in vertical directionA inputs for sample image, and * is two-dimensional convolution computing.
Preferably, union is taken to obtain being based on space block the local binary pattern operator with different radii and sampling number
The feature of the local binary patterns of shape information：
Wherein, LBP is local binary patterns characteristic value, and P is the number of neighbor pixel, and R is radius.
Preferably, the queue shape is fitted using the least square polynomial curve fitting method based on RANSAC bent
Line.
In order to achieve the above objects and other related objects, the present invention provides a kind of queue length control method, the queuing
Length control method at least includes：
Supervising device is set in the occasion for the length control that needs to rank, pedestrian's queuing video is gathered；
Queue length is detected using above-mentioned queue length automatic testing method；
When queue length exceeds default number, warning information is sent to client, client adjustment queue queue solves
Wait in line problem.
Preferably, the method for client adjustment queue queue includes：Adjust service window, the statistics volume of the flow of passengers or guiding pedestrian
Queue up.
As described above, the queue length automatic testing method and queue length control method of the present invention, with following beneficial
Effect：
The queue length automatic testing method and queue length control method of the present invention can fast and accurately detect monitoring
Pedestrian's queue length in scene, it is to avoid the problem of target is lost, reduces missing inspection；By monitoring queuing situation in real time, adjust in time
It is whole to handle window, so that the problem of effectively solving to wait in line, preferably serves the passengers.
Brief description of the drawings
Fig. 1 is shown as the schematic flow sheet of the queue length automatic testing method of the present invention.
Fig. 2~Fig. 4 is shown as the schematic diagram of three local binary patterns features of the present invention.
Fig. 5 is shown as the 8*8 patch schematic diagrames in the local binary patterns algorithm based on the block information in space of the present invention.
Fig. 6 is shown as the histogram of the pixel in the local binary patterns algorithm based on the block information in space of the present invention
Schematic diagram.
The number that Fig. 7 is shown as in the queue of the present invention calculates principle schematic.
Fig. 8 is shown as the schematic diagram of the queue length control method of the present invention.
Component label instructions
R1~r3 sample radius
S1~S6 steps
Embodiment
Illustrate embodiments of the present invention below by way of specific instantiation, those skilled in the art can be by this specification
Disclosed content understands other advantages and effect of the present invention easily.The present invention can also pass through specific realities different in addition
The mode of applying is embodied or practiced, the various details in this specification can also based on different viewpoints with application, without departing from
Various modifications or alterations are carried out under the spirit of the present invention.
Refer to Fig. 1~Fig. 7.It should be noted that the diagram provided in the present embodiment only illustrates this in a schematic way
The basic conception of invention, then in schema only display with relevant component in the present invention rather than according to package count during actual implement
Mesh, shape and size are drawn, and kenel, quantity and the ratio of each component can be a kind of random change during its actual implementation, and its
Assembly layout kenel may also be increasingly complex.
As shown in figure 1, the present invention provides a kind of queue length automatic testing method, comprise the following steps：
Step S1：Obtain pedestrian's queuing video, pedestrian's queuing video described in typing.
Specifically, the video that pedestrian queues up is obtained from monitoring system, by pedestrian's queuing video typing present invention's
In detecting system.
Step S2：Pedestrian in pedestrian's queuing video is detected based on organization of human body model algorithm.
Specifically, the pedestrian detection algorithm based on organization of human body model is built using the method based on shape.In this implementation
In example, pedestrian is recognized by constructing manikin.More specifically, constructing a two dimension based on pedestrian's part according to organization of human body
Model, is matched by the architectural feature (structoure of the human body) for extracting pedestrian image with two dimensional model；If the pedestrian's figure extracted
The part-structure feature organization of human body Model Matching corresponding with two dimensional model of picture, then detect pedestrian；If the row extracted
The part-structure feature organization of human body unmatched models corresponding with two dimensional model of people's image, then be not detected by pedestrian.The party
Method can effectively handle occlusion issue, it is possible to be inferred to the posture of human body.
Step S3：Detect after pedestrian, feature based on-line selection boosting algorithm constantly updates pedestrian image feature, with right
Pedestrian is tracked positioning.Previous frame is effectively solved by on-line selection boosting algorithm and detects target, and present frame target is lost
The problem of mistake, reduce missing inspection.
Strong classifier in the online boosting algorithm includes N number of feature selectorEach feature selecting
Device shares a feature pool, and the feature pool includes M feature hm, (m=1 ..., M), the M feature represent weak typing
Device.The online boosting algorithm maintains and updated the correct sample weights of accumulation classification of each feature alwaysWith classification error sample
This weightsThe tracking to pedestrian is realized with this.
Specifically, after target positioning, using the target area in tracking window as positive sample, by outside tracking window and mesh
Some background areas of the sizes such as mark are used as negative sample.As new samples (x, y), when y ∈ { 0,1 } are reached, N number of feature selectorSequentially generate, generation is to M feature h every timem, (m=1 ..., M) is updated, wherein, new samples
(x, y) is target area (positive sample) in current frame image in tracking window and tracking window is outer and the size such as target area
Some background areas (negative sample), x is the video in window with the size such as target area, and y is used to represent positive sample or negative sample, y
=0 is negative sample, and y=1 is positive sample：
Work as hm(x)=y when,
Work as hm(x) ≠ y when,
Wherein, λ is the current weights of sample, and initial value is 1.
After renewal is finished, each feature selector will select the feature of current minimum cumulative error frequency corresponding weak point as its
Class device：(m+The feature of i.e. current minimum cumulative error frequency), wherein, εmFor each feature
Cumulative error frequency, wherein, the feature n corresponding weight of cumulative error frequency is εnIt is characterized n pairs
The cumulative error frequency answered,Current minimum cumulative error frequency.
N-th of feature selectorAfter generation the weights λ of sample (x, y) also according to whether by mistake point being increased or
Reduce.Strong classifier is combined by this select feature according to respective weight after n times are selected：
Target new position in the environs of target location in previous frame with strong classifier evaluation by being determined.If searching for model
Enclose for S, the target candidate position in the S of hunting zone is pw(w=1 ..., W), target new position is xnew, thenIts
In, w+By the maximum position candidate of the overall merit of N number of strong classifier, to meet following formula：
Wherein,It is p i.e. in position candidatewFeature n feature selector is passed through at placeValue, can be with from above formula
N number of feature selector is needed to be evaluated in W position candidate when finding out target positioning.
Specifically, in the present embodiment, the feature of the on-line selection boosting algorithm includes：Haar features, edge direction
The feature of histogram feature and local binary patterns based on the block information in space, generates in target overall region, passes through three
The image with obvious rectangular characteristic, edge image and global texture are identified respectively for the calculating for planting feature, reduce single
Error and mistake that characteristic strip comes.
More specifically, Haar features are divided into four classes：Edge feature, linear character, central feature and diagonal feature, this four
Category feature is combined into feature templates.There are white and two kinds of rectangles of black in feature templates, and it is white to define the characteristic value of the template
Color rectangular pixels and subtract black rectangle pixel and.Haar characteristic values reflect the grey scale change situation of image.For example：Face
Some features can simply be described by rectangular characteristic, such as：Eyes are deeper than cheek color, and bridge of the nose both sides will than bridge of the nose color
Deep, face is deeper etc. than ambient color.In the present embodiment, Haar features are calculated using integration histogram, only traveled through
Image can greatly improve the efficiency of image feature value calculating to obtain the sums of all area pixels in image.
The main thought of integration histogram is that image is arrived to the formed rectangular area pixel sum of each point from the off to be used as one
The element of individual array is stored in internal memory, when to calculate the pixel and Shi Ke in some region with the element of direct index array, no
With recalculate this region pixel and, so as to accelerate calculating.Integration histogram can be under a variety of yardsticks, using identical
Time (constant time) calculate different features, therefore substantially increase detection speed.By substantial amounts of with relatively brighter
The subject image of aobvious Haar features (rectangle) trains grader with the method for pattern-recognition, and grader is a cascade, often
Level all retains the candidate's object with object features for entering next stage with the discrimination that is roughly the same, and the often subclassification of one-level
Device is then made up of many Haar features and (is calculated and obtained by integration histogram, and preserve lower position), have it is horizontal, vertical, incline
Oblique, and each one threshold value of characteristic strip and two branch values, one total threshold value of every grade of sub-classifier band.Recognize object
When, the same integration histogram that calculates is that subsequent calculations Haar features are prepared, and has object when then using with training
An equal amount of window traversal entire image of window, gradually amplifies window, equally does traversal search object later；Whenever window is moved
Move to a position, that is, calculate the Haar features in the window, compared after weighting with the threshold value of Haar features in grader so as to
A selection left side or right branch value, the branch value for the level that adds up are compared with the threshold value of corresponding stage, can just passed through more than the threshold value
Into next round screening.
More specifically, the extraction process of edge orientation histogram feature is as follows：If sample image input is A, edge is utilized
Operator extracts gradient magnitude for G in the horizontal directionx, the amplitude of gradient is G in vertical directiony, calculation formula is as follows, wherein *
For two-dimensional convolution computing：
It is defined on the amplitude GM of gradient on this pixelxyMeet following relation：
Next direction θ for defining the pixel gradient meets following relation：
Finally by the direction θ of the pixel gradient interval θ ∈ (- π, π] in be quantified as NCIndividual angular regions, and to region
The gradient of interior each pixel is counted, and corresponding amplitude is added into corresponding angle and histogram is obtained.It is every in histogram
The interval characteristic value of individual statistics is the edge feature that we need, and is collected into whole characteristic vector.
More specifically, because textural characteristics by illumination etc. are influenceed smaller, therefore selection local binary patterns are used as row
The Local textural feature in people region.Local binary patterns are obtained by analyzing the relation between pixel and its surrounding pixel point
Characteristic value.For adjacent pixel, the local binary patterns of central pixel are calculated by the following method：By phase
The numerical value g of adjacent pixelpWith the gray value g of central pixel pointcIt is compared, remembers if the gray value more than central pixel point
For 1,0 is designated as if the gray value less than central pixel point, vice versa.Hereafter using the position of a certain fixation as starting point, with
The two-value data that link is obtained clockwise obtains the two-value data string that length is eight, is translated into decimal data,
Just the local binary patterns characteristic value of the pixel is obtained.Computational methods are as follows：
Wherein, LBP is local binary patterns characteristic value, and P is the number of neighbor pixel, and R is radius, gpFor adjacent picture
The numerical value of vegetarian refreshments, gcCentered on pixel gray value.
Local binary patterns describe the local grain information of pedestrian well, but can not describe the global letter of pedestrian
Pattern on breath, such as pedestrian's clothes can make pedestrian detection precise decreasing.Further, propose in the present embodiment based on space
The local binary patterns algorithm of block information, in the present embodiment, Patch size is set as 8*8, in actual use, can basis
Algorithm requires the size of the setting patch, is not limited with the present embodiment.
As shown in Figure 2 to 4, in the present embodiment, by 3 local binary patterns with different radii and sampling number
Operator connects the feature L for obtaining the local binary patterns based on the block information in space.As shown in Fig. 2 the first local binary
The sample radius of pattern feature is r1, and sampling number is 8；As shown in figure 3, the sample radius of the second local binary patterns feature
For r2, sampling number is 12；As shown in figure 4, the sample radius of the 3rd local binary patterns feature is r3, sampling number is 16
It is individual；Wherein r1<r2<r3.Each point for image is allowd to obtain 3 different local binary patterns feature LBPP,R, this
The union of a little features contains the most important texture information of image：
As shown in figure 5, calculating the gradient of each pixel in the boxed area of patch, and calculate amplitude and the side of gradient
To, it is consistent with the amplitude of calculating gradient in edge orientation histogram feature and the method in direction, do not repeat one by one herein.Such as Fig. 6
It is shown, the statistics with histogram of the pixel of 8 directions (0 to 360 degree, every 45 degree of directions), Nogata in statistics boxed area
Figure statistical result is characterized vector (characteristic vectors of 8 dimensions).
Step S4：According to the pedestrian detected, the queue pattern curve of pedestrian is fitted.
Specifically, queue pattern curve is fitted using the least square polynomial curve fitting method based on RANSAC.
RANSAC algorithms (Random Sample Consensus, random sampling unification algorism) are to include abnormal data according to one group
Sample data set, calculates the mathematical model parameter of data, obtains the algorithm of effective sample data.It in 1981 by
Fischler and Bolles are proposed at first.The basic assumption of RANSAC algorithms is that comprising correct data, (inliers, can in sample
Data to be described by model), also comprising abnormal data, (outliers, deviation normal range (NR) is far, can not adapt to mathematical modeling
Data), i.e., contain noise in data set.These abnormal datas are probably the measurement due to mistake, the hypothesis of mistake, mistake
The generations such as calculating.RANSAC is also assumed that simultaneously, gives one group of correct data, exists to calculate and meet these data
The method of model parameter.RANSAC basic thoughts are described as follows：1. (b is first to the model that the minimum sampling cardinality of consideration one is b
Smallest sample number needed for beginningization model parameter) and a sample set Q, set Q sample number # (Q)>B, is randomly selected from Q
The subset T initialization models D of Q comprising b sample；2. complementary set TC=Q error in T with model D be less than a certain given threshold
T sample set and T constitutes T*.T* is considered interior point set, and they constitute T consistent collection (Consensus Set)；If 3. #
(T*) >=B, it is believed that obtain correct model parameter, and the methods such as least square are used again using T* (interior point inliers) is collected
Calculate new model D*；Again new T is randomly selected, above procedure is repeated.4. after certain frequency in sampling is completed, if not looking for
To consistent collection, then algorithm fails, and otherwise chooses the consistent collection of the maximum obtained after sampling and judges interior exterior point, algorithm terminates.
Step S5：Image calibration, world coordinates is converted to by image coordinate.
Step S6：The distance between queue physical length and adjacent pedestrian are calculated according to queue pattern curve, so as to calculate
Obtain the number in queue.
As shown in fig. 7, according to curve matching calculated curve length, that is, the length L queued up, according to the length L of queuing and team
The distance between phase neighbour d in row, calculates the number in queue, and calculation formula is as follows：
Number=L/d.
The present invention can carry out real-time, effective, quick and accurate under different application scene, Different climate and illumination condition
Pedestrian's queue length detection.
As shown in figure 8, the present invention also provides a kind of queue length control method, the queue length control method is at least wrapped
Include：
Supervising device is set in the occasion for needing queue length to control, pedestrian's queuing video is gathered.
Specifically, in the occasion of length control that needs to rank, the PE such as airport, hospital sets monitoring
The supervising devices such as probe.
Queue length is detected using above-mentioned queue length automatic testing method.
Specifically, pedestrian is detected based on organization of human body model algorithm；Using feature on-line selection boosting algorithm tracking pedestrians,
Mainly solve that previous frame detects target and present frame target reduces missing inspection, feature on-line selection boosting algorithm the problem of lose
Feature use Haar features, the spy of edge orientation histogram feature and the local binary patterns based on the block information in space
Levy, method scene according to residing for target of online selected characteristic chooses the expressive feature of tool automatically can improve tracking effect
Really；According to the people detected, the queue shape of people is fitted using the least square polynomial curve fitting method based on RANSAC,
So as to estimate the length of queue；Realize that image coordinate is tied to the conversion of world coordinate system by image calibration, so as to calculate reality
The length of queuing；According to curve matching calculated curve length, that is, the length queued up is adjacent in the length and queue further according to queuing
The distance between people, calculates the number in queue.Specific method is seen above, and is not repeated one by one herein.
When queue length exceeds default number, warning information is sent to client, client adjustment queue queue solves
Wait in line problem.
Specifically, the method for client adjustment queue queue includes：Operation management center is received after warning information, in time
Increase service window, alleviate the time waited in line；Client is counted to pedestrian's flow；Staff or early warning indicate and
When guiding pedestrian queue up, the pedestrian in the queue more than people is guided to the few queue of people；So as to effectively solve what is waited in line
Problem, is preferably served the passengers.
The present invention can monitor the queuing situation in check in region and security check region in real time, in time adjustment queuing team
Row, the problem of effectively solving to wait in line.
In summary, the present invention provides a kind of queue length automatic testing method and queue length control method, based on people
Body structural model algorithm detects pedestrian；Using feature on-line selection boosting algorithm tracking pedestrians, mainly solve previous frame and detect
The problem of target and present frame target are lost, reduces missing inspection, and it is special that the feature of feature on-line selection boosting algorithm uses Haar
Levy, the feature of edge orientation histogram feature and the local binary patterns based on the block information in space, the side of online selected characteristic
Method scene according to residing for target chooses the expressive feature of tool automatically can improve tracking effect；According to the people detected, adopt
The queue shape of people is fitted with the least square polynomial curve fitting method based on RANSAC, so as to estimate the length of queue；
Realize that image coordinate is tied to the conversion of world coordinate system by image calibration, so as to calculate the length of actual queuing；According to curve
The distance between phase neighbour in the Fitting Calculation length of curve, that is, the length queued up, the length and queue further according to queuing, calculates team
Number in row.The queue length automatic testing method and queue length control method of the present invention can be detected fast and accurately
Pedestrian's queue length in monitoring scene, it is to avoid the problem of target is lost, reduces missing inspection；By monitoring queuing situation in real time, and
When adjustment handle window, so as to effectively solve the problem of wait in line, preferably serve the passengers.So, effective gram of the present invention
Take various shortcoming of the prior art and have high industrial utilization.
The above-described embodiments merely illustrate the principles and effects of the present invention, not for the limitation present invention.It is any ripe
Know the personage of this technology all can carry out modifications and changes under the spirit and scope without prejudice to the present invention to above-described embodiment.Cause
This, those of ordinary skill in the art is complete without departing from disclosed spirit and institute under technological thought such as
Into all equivalent modifications or change, should by the present invention claim be covered.
Claims (14)
1. a kind of queue length automatic testing method, it is characterised in that the queue length automatic testing method at least includes：
Step S1：Obtain pedestrian's queuing video, pedestrian's queuing video described in typing；
Step S2：Pedestrian in pedestrian's queuing video is detected based on organization of human body model algorithm；
Step S3：Detect after pedestrian, feature based on-line selection boosting algorithm constantly updates pedestrian image characteristic value, with to row
People is tracked positioning；
Step S4：According to the pedestrian detected, the queue pattern curve of pedestrian is fitted；
Step S5：Image calibration, world coordinates is converted to by image coordinate；
Step S6：The distance between physical length of queue and adjacent pedestrian are calculated according to the queue pattern curve, so as to count
Calculate the number obtained in queue.
2. queue length automatic testing method according to claim 1, it is characterised in that：Step S2 is specifically included：According to
Organization of human body constructs a two dimensional model based on pedestrian's part, is carried out by the architectural feature for extracting pedestrian image with two dimensional model
Matching, so as to recognize pedestrian.
3. queue length automatic testing method according to claim 1, it is characterised in that：The online boosting algorithm enters one
Step includes：N number of feature selector, is used as strong classifier；M feature constitutes a feature pool, is used as Weak Classifier；When new samples are arrived
Up to when, N number of feature selector is sequentially generated, every time accumulation classification correct sample weights and iterated integral of the generation to M feature
Class error sample weights are updated, and each feature selector is using the feature of current minimum cumulative error frequency corresponding weak point as its
Class device, N number of feature selector combines to form strong classifier, and target new position is by previous frame in the environs of target location
Determined with strong classifier evaluation.
4. queue length automatic testing method according to claim 3, it is characterised in that：Accumulation classification to M feature
The method that correct sample weights and accumulation classification error sample weights are updated is specifically included：
Work as hm(x)=y when,
<mrow>
<msubsup>
<mi>&lambda;</mi>
<mi>m</mi>
<mrow>
<mi>c</mi>
<mi>o</mi>
<mi>r</mi>
<mi>r</mi>
</mrow>
</msubsup>
<mo>=</mo>
<msubsup>
<mi>&lambda;</mi>
<mi>m</mi>
<mrow>
<mi>c</mi>
<mi>o</mi>
<mi>r</mi>
<mi>r</mi>
</mrow>
</msubsup>
<mo>+</mo>
<mi>&lambda;</mi>
</mrow>
Work as hm(x) ≠ y when,
<mrow>
<msubsup>
<mi>&lambda;</mi>
<mi>m</mi>
<mrow>
<mi>w</mi>
<mi>r</mi>
<mi>o</mi>
<mi>n</mi>
<mi>g</mi>
</mrow>
</msubsup>
<mo>=</mo>
<msubsup>
<mi>&lambda;</mi>
<mi>m</mi>
<mrow>
<mi>w</mi>
<mi>r</mi>
<mi>o</mi>
<mi>n</mi>
<mi>g</mi>
</mrow>
</msubsup>
<mo>+</mo>
<mi>&lambda;</mi>
</mrow>
Wherein, hm, (m=1 ..., M) is characterized, and x is the video in window with the size such as target area, y be used to representing positive sample or
Negative sample, y=0 is negative sample, and y=1 is positive sample,To accumulate the correct sample weights of classification,It is wrong for accumulation classification
Sample weights are missed, λ is the current weights of sample, and initial value is 1.
5. queue length automatic testing method according to claim 3, it is characterised in that：The strong classifier is by N number of spy
Selector is levied to combine to be formed by weight：
<mrow>
<mi>H</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>s</mi>
<mi>i</mi>
<mi>g</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>n</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>N</mi>
</munderover>
<msub>
<mi>&alpha;</mi>
<mi>n</mi>
</msub>
<msubsup>
<mi>h</mi>
<mi>n</mi>
<mrow>
<mi>s</mi>
<mi>e</mi>
<mi>l</mi>
</mrow>
</msubsup>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
<mo>)</mo>
</mrow>
</mrow>
Weight αnMeet following relation：
<mrow>
<msub>
<mi>&alpha;</mi>
<mi>n</mi>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<mi>l</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>&epsiv;</mi>
<mi>n</mi>
</msub>
</mrow>
<msub>
<mi>&epsiv;</mi>
<mi>n</mi>
</msub>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
Wherein,It is characterized selector, εnIt is characterized the corresponding cumulative error frequencies of n.
6. queue length automatic testing method according to claim 5, it is characterised in that：Target new position meets such as ShiShimonoseki
System：
<mrow>
<msub>
<mi>x</mi>
<mrow>
<mi>n</mi>
<mi>e</mi>
<mi>w</mi>
</mrow>
</msub>
<mo>=</mo>
<msub>
<mi>p</mi>
<msup>
<mi>w</mi>
<mo>+</mo>
</msup>
</msub>
<mo>,</mo>
</mrow>
1
Wherein,pw(w=1 ..., W) is the target candidate position in hunting zone,It is characterized selector, αnFor weight.
7. the queue length automatic testing method according to claim 1 or 3, it is characterised in that：The online boosting algorithm
In feature include：The spy of Haar features, edge orientation histogram feature and the local binary patterns based on the block information in space
Levy.
8. queue length automatic testing method according to claim 7, it is characterised in that：The calculating side of the Haar features
Method uses integration histogram.
9. queue length automatic testing method according to claim 7, it is characterised in that：The edge orientation histogram is special
The computational methods levied are as follows：By the direction θ of pixel gradient interval θ ∈ (- π, π] in be quantified as multiple angular regions, it is and right
The gradient of each pixel is counted in region, and corresponding amplitude is added into corresponding angle and histogram is obtained.
10. queue length automatic testing method according to claim 9, it is characterised in that：The side of the pixel gradient
To meeting following relation：
<mrow>
<mi>&theta;</mi>
<mo>=</mo>
<mi>arctan</mi>
<mrow>
<mo>(</mo>
<mfrac>
<msub>
<mi>G</mi>
<mi>y</mi>
</msub>
<msub>
<mi>G</mi>
<mi>x</mi>
</msub>
</mfrac>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
Wherein, gradient magnitude is extracted in the horizontal directionThe amplitude of gradient in vertical directionA inputs for sample image, and * is two-dimensional convolution computing.
11. queue length automatic testing method according to claim 1, it is characterised in that：There to be different radii and adopt
The local binary pattern operator of number of samples takes union to obtain the features of the local binary patterns based on the block information in space：
<mrow>
<mi>L</mi>
<mo>=</mo>
<munder>
<mrow>
<mi></mi>
<mo>&cup;</mo>
</mrow>
<mrow>
<mi>P</mi>
<mo>,</mo>
<mi>R</mi>
</mrow>
</munder>
<msub>
<mi>LBP</mi>
<mrow>
<mi>P</mi>
<mo>,</mo>
<mi>R</mi>
</mrow>
</msub>
</mrow>
Wherein, LBP is local binary patterns characteristic value, and P is the number of neighbor pixel, and R is radius.
12. queue length automatic testing method according to claim 1, it is characterised in that：Using based on RANSAC most
A young waiter in a wineshop or an inn multiplies polynomial curve fitting method to be fitted the queue pattern curve.
13. a kind of queue length control method, it is characterised in that the queue length control method at least includes：
Supervising device is set in the occasion for the length control that needs to rank, pedestrian's queuing video is gathered；
Queue length is detected using the queue length automatic testing method as described in claim 1~12 any one；
When queue length exceeds default number, warning information is sent to client, client adjustment queue queue solves to queue up
Waiting problem.
14. queue length control method according to claim 13, it is characterised in that：Client adjusts the side of queue queue
Method includes：Service window, the statistics volume of the flow of passengers or guiding pedestrian is adjusted to queue up.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710312890.3A CN107153819A (en) | 2017-05-05 | 2017-05-05 | A kind of queue length automatic testing method and queue length control method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710312890.3A CN107153819A (en) | 2017-05-05 | 2017-05-05 | A kind of queue length automatic testing method and queue length control method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107153819A true CN107153819A (en) | 2017-09-12 |
Family
ID=59794055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710312890.3A Pending CN107153819A (en) | 2017-05-05 | 2017-05-05 | A kind of queue length automatic testing method and queue length control method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107153819A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110517399A (en) * | 2019-08-30 | 2019-11-29 | 山东浪潮商用系统有限公司 | A kind of realization method and system for tax handling service hall's business intelligence early warning |
| CN110555363A (en) * | 2018-05-30 | 2019-12-10 | 佳能株式会社 | Information processing apparatus, image pickup apparatus, information processing method, and storage medium |
| CN111008611A (en) * | 2019-12-20 | 2020-04-14 | 浙江大华技术股份有限公司 | Queuing time determining method and device, storage medium and electronic device |
| CN111325057A (en) * | 2018-12-14 | 2020-06-23 | 杭州海康威视数字技术股份有限公司 | Queuing queue detection method and device |
| WO2020140749A1 (en) * | 2019-01-02 | 2020-07-09 | 京东方科技集团股份有限公司 | Queuing recommendation method and device, terminal, and computer readable storage medium |
| CN112288792A (en) * | 2020-09-28 | 2021-01-29 | 上海数川数据科技有限公司 | Vision-based instant measurement method for guest queuing length and waiting time |
| CN112735019A (en) * | 2021-01-20 | 2021-04-30 | 联仁健康医疗大数据科技股份有限公司 | Queuing guidance method, device, system, electronic equipment and storage medium |
| CN112954268A (en) * | 2019-12-10 | 2021-06-11 | 晶睿通讯股份有限公司 | Queue analysis method and image monitoring equipment |
| CN112949350A (en) * | 2019-12-10 | 2021-06-11 | 晶睿通讯股份有限公司 | Queue analysis method and image monitoring equipment thereof |
| CN117576822A (en) * | 2023-11-20 | 2024-02-20 | 上海徽视科技集团有限公司 | Queuing and number calling guiding system based on Internet platform |
| CN117576822B (en) * | 2023-11-20 | 2024-04-30 | 上海徽视科技集团有限公司 | Queuing and number calling guiding system based on Internet platform |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103761515A (en) * | 2014-01-27 | 2014-04-30 | 中国科学院深圳先进技术研究院 | Human face feature extracting method and device based on LBP |
| CN103903445A (en) * | 2014-04-22 | 2014-07-02 | 北京邮电大学 | Vehicle queuing length detection method and system based on video |
| CN103971095A (en) * | 2014-05-09 | 2014-08-06 | 西北工业大学 | Large-scale facial expression recognition method based on multiscale LBP and sparse coding |
| CN104134078A (en) * | 2014-07-22 | 2014-11-05 | 华中科技大学 | Automatic selection method for classifiers in people flow counting system |
| CN104835142A (en) * | 2015-03-10 | 2015-08-12 | 杭州电子科技大学 | Vehicle queuing length detection method based on texture features |
| US20150312529A1 (en) * | 2014-04-24 | 2015-10-29 | Xerox Corporation | System and method for video-based determination of queue configuration parameters |
| CN105447492A (en) * | 2015-11-13 | 2016-03-30 | 重庆邮电大学 | Image description method based on 2D local binary pattern |
| US20160180173A1 (en) * | 2014-12-18 | 2016-06-23 | Sensormatic Electronics, LLC | Method and System for Queue Length Analysis |
| CN105763853A (en) * | 2016-04-14 | 2016-07-13 | 北京中电万联科技股份有限公司 | Emergency early warning method for stampede accident in public area |
| CN106096553A (en) * | 2016-06-06 | 2016-11-09 | 合肥工业大学 | A kind of pedestrian traffic statistical method based on multiple features |
| CN106128121A (en) * | 2016-07-05 | 2016-11-16 | 中国石油大学(华东) | Vehicle queue length fast algorithm of detecting based on Local Features Analysis |
| CN106250820A (en) * | 2016-07-20 | 2016-12-21 | 华南理工大学 | A kind of staircase mouth passenger flow based on image procossing is blocked up detection method |
-
2017
- 2017-05-05 CN CN201710312890.3A patent/CN107153819A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103761515A (en) * | 2014-01-27 | 2014-04-30 | 中国科学院深圳先进技术研究院 | Human face feature extracting method and device based on LBP |
| CN103903445A (en) * | 2014-04-22 | 2014-07-02 | 北京邮电大学 | Vehicle queuing length detection method and system based on video |
| US20150312529A1 (en) * | 2014-04-24 | 2015-10-29 | Xerox Corporation | System and method for video-based determination of queue configuration parameters |
| CN103971095A (en) * | 2014-05-09 | 2014-08-06 | 西北工业大学 | Large-scale facial expression recognition method based on multiscale LBP and sparse coding |
| CN104134078A (en) * | 2014-07-22 | 2014-11-05 | 华中科技大学 | Automatic selection method for classifiers in people flow counting system |
| US20160180173A1 (en) * | 2014-12-18 | 2016-06-23 | Sensormatic Electronics, LLC | Method and System for Queue Length Analysis |
| CN104835142A (en) * | 2015-03-10 | 2015-08-12 | 杭州电子科技大学 | Vehicle queuing length detection method based on texture features |
| CN105447492A (en) * | 2015-11-13 | 2016-03-30 | 重庆邮电大学 | Image description method based on 2D local binary pattern |
| CN105763853A (en) * | 2016-04-14 | 2016-07-13 | 北京中电万联科技股份有限公司 | Emergency early warning method for stampede accident in public area |
| CN106096553A (en) * | 2016-06-06 | 2016-11-09 | 合肥工业大学 | A kind of pedestrian traffic statistical method based on multiple features |
| CN106128121A (en) * | 2016-07-05 | 2016-11-16 | 中国石油大学(华东) | Vehicle queue length fast algorithm of detecting based on Local Features Analysis |
| CN106250820A (en) * | 2016-07-20 | 2016-12-21 | 华南理工大学 | A kind of staircase mouth passenger flow based on image procossing is blocked up detection method |
Non-Patent Citations (5)
| Title |
|---|
| 刘中华等: "基于多尺度局部二值模式的人脸识别", 《计算机科学》 * |
| 查宇飞等: "《视频目标跟踪方法》", 31 July 2015 * |
| 涂宏斌等著: "《基于机器学习的行为识别技术研究》", 30 September 2016 * |
| 邱天圆: "基于多特征融合的行人检测算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 颜佳等: "遮挡环境下采用在线Boosting的目标跟踪", 《光学精密工程》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110555363A (en) * | 2018-05-30 | 2019-12-10 | 佳能株式会社 | Information processing apparatus, image pickup apparatus, information processing method, and storage medium |
| CN111325057B (en) * | 2018-12-14 | 2024-02-27 | 杭州海康威视数字技术股份有限公司 | Queuing queue detection method and device |
| CN111325057A (en) * | 2018-12-14 | 2020-06-23 | 杭州海康威视数字技术股份有限公司 | Queuing queue detection method and device |
| WO2020140749A1 (en) * | 2019-01-02 | 2020-07-09 | 京东方科技集团股份有限公司 | Queuing recommendation method and device, terminal, and computer readable storage medium |
| US11488321B2 (en) * | 2019-01-02 | 2022-11-01 | Beijing Boe Technology Development Co., Ltd. | Queuing recommendation method and device, terminal and computer readable storage medium |
| CN110517399A (en) * | 2019-08-30 | 2019-11-29 | 山东浪潮商用系统有限公司 | A kind of realization method and system for tax handling service hall's business intelligence early warning |
| CN112949350A (en) * | 2019-12-10 | 2021-06-11 | 晶睿通讯股份有限公司 | Queue analysis method and image monitoring equipment thereof |
| CN112954268A (en) * | 2019-12-10 | 2021-06-11 | 晶睿通讯股份有限公司 | Queue analysis method and image monitoring equipment |
| US11138754B2 (en) | 2019-12-10 | 2021-10-05 | Vivotek Inc. | Queue analyzing method and image monitoring apparatus |
| TWI756597B (en) * | 2019-12-10 | 2022-03-01 | 晶睿通訊股份有限公司 | Queue analyzing method and image monitoring apparatus |
| CN112954268B (en) * | 2019-12-10 | 2023-07-18 | 晶睿通讯股份有限公司 | Queue analysis method and image monitoring equipment |
| CN112949350B (en) * | 2019-12-10 | 2024-04-09 | 晶睿通讯股份有限公司 | Queue analysis method and image monitoring device thereof |
| CN111008611A (en) * | 2019-12-20 | 2020-04-14 | 浙江大华技术股份有限公司 | Queuing time determining method and device, storage medium and electronic device |
| CN112288792A (en) * | 2020-09-28 | 2021-01-29 | 上海数川数据科技有限公司 | Vision-based instant measurement method for guest queuing length and waiting time |
| CN112288792B (en) * | 2020-09-28 | 2022-12-02 | 上海数川数据科技有限公司 | Vision-based method for measuring guest queuing length and waiting time in real time |
| CN112735019A (en) * | 2021-01-20 | 2021-04-30 | 联仁健康医疗大数据科技股份有限公司 | Queuing guidance method, device, system, electronic equipment and storage medium |
| CN117576822A (en) * | 2023-11-20 | 2024-02-20 | 上海徽视科技集团有限公司 | Queuing and number calling guiding system based on Internet platform |
| CN117576822B (en) * | 2023-11-20 | 2024-04-30 | 上海徽视科技集团有限公司 | Queuing and number calling guiding system based on Internet platform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107153819A (en) | A kind of queue length automatic testing method and queue length control method | |
| CN110738101B (en) | Behavior recognition method, behavior recognition device and computer-readable storage medium | |
| CN109934115B (en) | Face recognition model construction method, face recognition method and electronic equipment | |
| CN109635875A (en) | A kind of end-to-end network interface detection method based on deep learning | |
| CN110084161A (en) | A kind of rapid detection method and system of skeleton key point | |
| CN108830188A (en) | Vehicle checking method based on deep learning | |
| CN110188720A (en) | A kind of object detection method and system based on convolutional neural networks | |
| CN105205453B (en) | Human eye detection and localization method based on depth self-encoding encoder | |
| CN108805016A (en) | A kind of head and shoulder method for detecting area and device | |
| CN109029363A (en) | A kind of target ranging method based on deep learning | |
| CN105069751B (en) | A kind of interpolation method of depth image missing data | |
| CN107705322A (en) | Motion estimate tracking and system | |
| CN106934795A (en) | The automatic testing method and Forecasting Methodology of a kind of glue into concrete beam cracks | |
| CN109558902A (en) | A kind of fast target detection method | |
| CN109271990A (en) | A kind of semantic segmentation method and device for RGB-D image | |
| CN104732546B (en) | The non-rigid SAR image registration method of region similitude and local space constraint | |
| CN109145836A (en) | Ship target video detection method based on deep learning network and Kalman filtering | |
| CN109740454A (en) | A kind of human body posture recognition methods based on YOLO-V3 | |
| Singh et al. | Human pose estimation using convolutional neural networks | |
| CN106874913A (en) | A kind of vegetable detection method | |
| CN108334878A (en) | Video images detection method and apparatus | |
| Öztürk et al. | Transfer learning and fine‐tuned transfer learning methods' effectiveness analyse in the CNN‐based deep learning models | |
| CN109271848A (en) | A kind of method for detecting human face and human face detection device, storage medium | |
| CN111626241B (en) | Face detection method and device | |
| CN114022554A (en) | Massage robot acupuncture point detection and positioning method based on YOLO |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170912 |