CN101980245A - Adaptive template matching-based passenger flow statistical method - Google Patents
Adaptive template matching-based passenger flow statistical method Download PDFInfo
- Publication number
- CN101980245A CN101980245A CN2010105092177A CN201010509217A CN101980245A CN 101980245 A CN101980245 A CN 101980245A CN 2010105092177 A CN2010105092177 A CN 2010105092177A CN 201010509217 A CN201010509217 A CN 201010509217A CN 101980245 A CN101980245 A CN 101980245A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- model
- video
- image
- 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.)
- Granted
Links
Images
Abstract
The invention discloses an adaptive template matching-based passenger flow statistical method. A front end and a rear end are adopted in the method, wherein the front end is connected with an image acquisition device by using an industrial personal computer and used for analysis and statistics of monitored video; the rear end is a PC (personal computer machine) and is used for processing the monitored video and the statistical result transmitted by the front end and storing the processing results to local; the front end constructs a plurality of weak classifiers by using a Haar characteristic-based Adaboost algorithm and connects the heads and the tails of the weak classifiers, and the output of one weak classifier is used as the input of the other classifier to obtain a cascaded strong classifier for single passenger detection; aiming at the passenger flow statistical characteristics, a sensitive area is set in the acquired digital video to reduce the processing complexity and improve the processing accuracy; after the cascaded strong classifier constructed by adopting the Haar characteristic-based Adaboost algorithm detects the passenger, modeling is performed, and the parts of the passenger body are endowed with different values; the passenger is tracked by adopting a template matching method to obtain a passenger moving track; and the template is updated after the template is changed so that the template has adaptive characteristic.
Description
Technical field
The present invention relates to a kind of passenger flow statistical method, belong to target following and mode identification technology, wherein relate to detection, identification and the tracking of the expression of target and modeling, target based on adaptive template coupling.
Background technology
For the energy savings and the consideration of raising the efficiency, when the deviser designs some zone big or small, usually need the statistics of earlier this regional volume of the flow of passengers being carried out, with this foundation as design.Each emporium is for the consideration that improves the sales volume, also may need the customer quantity of different periods in one day is added up, and formulates various sales tactics with this.At the underground railway track field of traffic, the method for passenger flow statistics mainly is to add up by the number of times that the passenger is swiped the card to obtain at present.Along with putting into operation of automatic ticket-selling system, certain time period pass in and out certain station the volume of the flow of passengers can by computer technology easily fast the statistics obtain, but if desired statistics be positioned at subway station swipe the card the zone with the passenger flow of some passages, just lack a kind of statistical method of efficiently and accurately.
Suppose and have two subway line L1 and L2, subway station S is intersection's (or being referred to as the transfer stop) of L1 and L2, there is the underpass T that communicates with one another between L1 and the L2, the passenger swipe the card or buy tickets enter website S after, may go to L2 through T from L1, also may go to L1 through T from L2, at this moment, if need add up to the volume of the flow of passengers in certain time period internal channel T, owing to do not need to swipe the card again during passenger's process passage T, therefore under this scene, calculate person-time method of adding up of swiping the card and just can't meet the demands.
Be applied at present that passenger flow statistical method under the above-mentioned scene mainly contains infrared photoelectric sensor statistics, laser statistics and by means of the video statistics method of computer technology.The advantage of infrared electro formula statistical method be easy for installation, be convenient to safeguard that the car door that has begun to be widely used in bus, Tube and Train is added up passenger number up and down.The major defect of infrared electro formula passenger number statistical system capable is the product that has only the section formula at present, and in addition, section width is bigger importing and exporting, under the crowded situation, because crowd that can't the split position next-door neighbour, and cause statistical precision to descend.The method of laser statistics makes its vertical scanning by laser scanner is set above passage or gateway.When the pedestrian passes through the laser curtain, convex closure can appear in the laser scanning data of present frame, reflect pedestrian's outline.The current number by laser scanning face pedestrian of peak bag number representative, the height representative of each peak bag is at pedestrian's height of this vertical scanning tangent plane.There is the shortcoming of the maximum identical with infrared electro formula statistical method in the laser statistical method, does not promptly have image recording in the statistics, and this has brought difficulty for observation of real-time monitored field condition and later stage scene reproduction.In order to overcome this difficulty, also be the demand of actual environment for use simultaneously, these two kinds of methods are often together used in conjunction with this video monitoring.
Because infrared photoelectric sensor statistics and laser statistical method all need video monitoring as supplementary means, therefore, if by means of computer technology, by the video image that collects being understood and being analyzed, the pedestrian's who obtains occurring in the scene quantity, cocoa reduces the complexity of total system greatly, also provides convenience for the various processing in later stage simultaneously.
Have more directly perceived, more convenient use, more efficient, compatibility is stronger characteristics by means of the video statistics method infrared photoelectric sensor of computer technology statistics and laser statistical method.The dispatching center can obtain the volume of the flow of passengers of certain time period in certain passage rapidly accurately; Pedestrian's video motion track can be obtained,, even pedestrian's face-image can be obtained according to actual conditions.
Summary of the invention
Technology of the present invention is dealt with problems: the deficiency that overcomes existing passenger flow statistical method, a kind of passenger flow statistical method based on the adaptive template coupling is provided, this system can add up the volume of the flow of passengers in each time period in certain monitoring scene fast and accurately, pedestrian's video motion track can be obtained, and pedestrian's face-image might be obtained.
Technical solution of the present invention: a kind of passenger flow statistical method based on the adaptive template coupling, this method is to realize by a kind of passenger flow statistical system based on the adaptation module coupling, this passenger flow statistical system is made up of front-end and back-end, front end uses an industrial computer to connect image capture device, and monitor video is analyzed and added up; The rear end is a PC, monitor video and statistics that each front end sends is handled, and be saved in this locality; System front end is handled and is comprised the steps: that (1) front end industrial computer obtains frame of video from image capture device, supports to handle simultaneously multi-channel video; (2) according to the characteristics of passenger flow statistics, in the frame of video of obtaining, set the sensitizing range, only these regional data are handled; (3) use Adaboost algorithm to make up a plurality of Weak Classifiers based on the Haar feature, and these Weak Classifiers are connected from beginning to end, the output of a Weak Classifier is as the input of another sorter, obtain the strong classifier of a cascade, in the sensitizing range of video image, carry out single pedestrian's detection, obtain single pedestrian's image; (4) single pedestrian's image that detection is obtained carries out modeling, actual characteristic when blocking between the pedestrian, give different weights to the various piece of pedestrian's human body, be arranged in the position of frame of video, position and area size and the colouring information that the pedestrian partes corporis humani divides in conjunction with this pedestrian's image, for this pedestrian sets up model; (5) method of employing template matches is mated each model after the modeling and existing model (if existence), if it fails to match, thinks that then the pedestrian of this model representative has just entered the sensitizing range; If the match is successful, show that the pedestrian of this model representative has appeared in the sensitizing range, upgrade the overall positions of this model and the weight of each several part, make template matches have adaptive characteristic; (6) check existing model, if in from the current video frame sensitizing range, detecting the pedestrian dummy that also modeling obtains, do not have the model that is complementary with it, think that then the pedestrian of this model representative has left the sensitizing range, check the motion path (position of pedestrian's image in the frame of video sensitizing range by the model representative characterizes) of this model this moment, and carry out the statistics of passenger flow.
Wherein, the sensitizing range described in the step (2) is the artificial rectangular area of setting of in the frame of video of being obtained in the step (1), and system front end is only handled the video data in this rectangular area.
The design of the strong classifier of the cascade described in the step (3) is by extracting single pedestrian's eigenwert and using the Adaboost algorithm to extract two parts of eigenwert design Weak Classifier and form, single pedestrian's feature of extracting has edge feature, symmetrical feature and central feature, and these features develop by the Haar feature; When using the Adaboost algorithm to extract eigenwert design Weak Classifier,, scan the central area computing center feature of the moving window of entire image earlier, then in the marginal portion of subwindow edge calculation feature according to single pedestrian's feature of image; After removing most of invalid rectangular characteristic according to these prioris, utilize the AdaBoost algorithm to pick out again to the best feature of single pedestrian's property distinguished, make up each Weak Classifier, when the false drop rate of Weak Classifier during less than setting threshold, stop to make up Weak Classifier, all Weak Classifiers that made up first places are linked to each other, obtain the strong classifier of a cascade, be used for the single pedestrian of sensitizing range detection.
Wherein, described step (4) is according to the position of human limb and the priori of structure, human body is divided into 3 parts, be respectively head, upper limbs and lower limb are given different weights to various piece, the actual conditions when blocking each other at pedestrian in the monitoring scene, give the highest weights to head, lower limb are given minimum weights; To detecting every the single pedestrian's image that obtains in the step (3), calculate the residing position in the sensitizing range of every head, upper limbs and lower limb in pedestrian's image and shared area size, and the color histogram feature of various piece, this pedestrian is carried out modeling, obtain characterizing this pedestrian's model.
Wherein, described step (5) is to adopt the method for template matches, model and the existing model set T that sets up in the step (4) is mated when initial (have the model number be 0); If it fails to match, think that then the pedestrian of this model representative has just entered the sensitizing range, this model is added set T; The match is successful, shows that the pedestrian of this model representative has appeared in the sensitizing range, upgrades the model of this model correspondence among the T.
Wherein, described step (6) is after handling all models of being set up in the step (4), check existing model set T, if having certain is not initiate model, this model is not updated simultaneously, think that then the pedestrian of this model representative has left the sensitizing range, check the motion path (position of pedestrian's image in the frame of video sensitizing range by the model representative characterizes) of this model this moment, carry out the renewal of the volume of the flow of passengers.
The present invention's advantage compared with prior art is:
(1) can add up accurately the volume of the flow of passengers in bus's flow monitoring scene, and can handle the situation of blocking between the pedestrian;
(2) can accurately obtain pedestrian's movement locus, and reflect with the form of video image;
(3) the anti-complex background interference capability of system is strong, can adapt to the acute variation of environment such as weather, illumination, and can accomplish real-time processing;
(4) system is made of front-end and back-end, and front end is responsible for monitor data is handled, and obtains result, and data are sent to the rear end; Tabulate statistics is carried out to the data that each front end transmits in the rear end;
Description of drawings
Fig. 1 is a system global structure synoptic diagram of the present invention;
Fig. 2 is the processing flow chart of front end of the present invention;
Fig. 3 is the coordinate system of front-end processing of the present invention;
The Haar-Like feature that Fig. 4 (1) (2) (3) proposes for the present invention; Wherein (1) is edge feature, and (2) are symmetrical feature, and (3) are central feature;
The tandem type sorter structure that Fig. 5 proposes for the present invention;
The organization of human body piecemeal synoptic diagram that Fig. 6 proposes for the present invention;
Embodiment
As shown in Figure 1, a kind of passenger flow statistical method based on the adaptive template coupling, this method is to realize by a kind of passenger flow statistical system based on the adaptation module coupling, this passenger flow statistical system is made up of front-end and back-end, each front end uses an industrial computer to connect image capture device, and monitor video is analyzed and added up; The rear end is a data server, monitor video and statistics that each front end sends is handled, and be saved in this locality.
The all operations of front end are all carried out in coordinate system as shown in Figure 3.F (0,0, W, the H) frame of video obtained of expression, the coordinate of frame of video F top left corner apex is (0,0), i.e. true origin, the coordinate on summit, the lower right corner be (W, H), W represents the width of frame of video, H represents the height of frame of video.Making that S1 is existing model set, be sky when initial.
As described in step 1 and step 2 in the summary of the invention, if there is the frame of video that has been untreated, then take out first frame, all model states among the S 1 are set to not upgrade.To the monitoring video frame of obtaining, at first in frame of video, delimit a rectangular area as the sensitizing range.R (L, T, W
r, H
r) represent that this sensitizing range, L and T represent the distance of top left corner apex range coordinate initial point on the X and Y coordinates axle of sensitizing range, W respectively
rThe width of expression sensitizing range, H
rThe height of expression sensitizing range.
As described in the step 3 in the summary of the invention,, at first carry out single pedestrian's detection to the video data in the sensitizing range.
For the consideration of robustness and real-time, be subjected to the inspiration of 4 kinds of Haar-like features of Viola-Jones algorithm proposition simultaneously, constructed a stack features single pedestrian's image has been carried out modeling.The principle of selecting feature is the reflection that essence can be arranged the spatial structure characteristic of single pedestrian's image, and is affected by noise little, and these features will be convenient to extract, and can calculate simply fast on computers.The rectangular characteristic of designing is divided into 3 kinds of edge features, symmetrical feature, central feature, as shown in Figure 4.
Calculate for convenience, introduced the notion of integral image:
The integral image values of each pixel that utilization calculates after the limited number of time computing, can calculate the rectangular characteristic value deducts 1 area grayscale value for the summation of-1 area grayscale value summation fast.Yet, the rectangular characteristic T of piece image
i, i is its feature number, the sum of i will be higher than the number of pixel far away, therefore, must have one efficiently sorter could filter out useful feature effectively, therefore adopt the sorter of tandem type to reach this purpose.A plurality of Weak Classifiers are coupled together the sorter that has just obtained cascade from beginning to end.Weak Classifier is meant that the classification performance of this sorter is bad, but the tandem type sorter that a plurality of Weak Classifier cascades obtain but has extraordinary classification performance.
N single pedestrian's sample of input and non-pedestrian's training sample: { x
1, y
1..., { x
n, y
n, wherein, y
i=true, and false}, i=1,2 ..., n.A dummy copy and b true sample arranged in the known training sample.J the Weak Classifier that feature generates, as the formula (2).
Wherein, h
iThe value of expression Weak Classifier; θ
iBe threshold value; p
iThe direction of the expression sign of inequality, can only get ± 1; f
i(x) representation feature value.
In the AdaBoost algorithm, constitute a strong classifier by several Weak Classifiers, its concrete steps are as follows:
(1) initialization error weight is for h
i=0 dummy copy, weight w
1, i=1/2a; For h
i=1 true sample, weight w
1, i=1/2b;
(2) select Weak Classifier, upgrade the error weight of each sample with least error:
Wherein, if i sample correctly classified e then
i=0, on the contrary e
i=i; β
t=ε
t/ (1-ε
t).
(3) choose T Weak Classifier h with minimum error
t(x) strong classifier of Xing Chenging is H (x).
After designing strong classifier, set of weights is synthesized a stacked sorter.Before the number of plies of determining cascade classifier and concrete parameter, determine the maximum false drop rate f of system earlier
MaxWith minimum detection rate d
MinIf the non-single pedestrian's of the correct identification of sorter ratio is B, the ratio that wherein B ∈ (0,1), and sorter mistake filters single pedestrian zone is Err, Err ∈ (0,1) wherein, the cascade classifier that is N at a level then, its cascade classifier minimum detection rate d
Min=(1-Err)
N, maximum false drop rate f
Max=(1-B)
NIn the training process of algorithm, best sorter is constantly added to come in, till reaching predefined false detection rate.
When the stacked sorter of the single pedestrian of training, need the positive negative example base of the incompatible structure of other background interference image sets except that the pedestrian in a large amount of single pedestrian's image collections and the practical application scene.Because single pedestrian's image is from different collecting devices, the picture quality difference comprises situations such as different illumination conditions, weather condition, shooting angle.Therefore, need carry out (comprising positive negative sample) the pixel normalizing words of image to the image pattern storehouse, the pixel value after the normalization is
Wherein: G is the gray-scale value of image slices vegetarian refreshments,
Be the average of image pixel, σ is the mean square deviation of image pixel.In order to improve the robustness of stacked sorter, training sample is wanted representative and popularity, and promptly positive sample will comprehensively reflect the rectangular characteristic in front part of vehicle front, and negative sample needs comprehensively to reflect the feature of all kinds of interference elements in the background.Concrete step is as follows:
(1) the positive and negative samples pictures of structure storehouse.
(2) each layer training Boosting training: from all features, choose the Weak Classifier of whole error rate minimum after the weighting, carry out weight by the training of sample and upgrade, up to reaching verification and measurement ratio and the false drop rate that needs.
(3) strong classifier of each layer training is together in series forms stacked sorter, be used for target detection.
Pedestrian's motion is a process that draws near, and pedestrian's characteristic information at a distance is also not obvious and cause the rising of false drop rate easily, therefore, in training, the window of positive sample is carried out normalization.Pedestrian's size of images is the continually varying process of changing from small to big in the continuous videos, in order to detect pedestrian's image of various sizes as far as possible, adopt the scan mode of image pyramid, image is carried out within the specific limits successively the convergent-divergent of size, with the input of scaled images as image to be checked, the output result is possible single pedestrian zone.
The structure of the sorter of the cascade of design as shown in Figure 5.Carry out the branch time-like at each layer, if current subwindow is detected as non-single pedestrian zone, then sorter stops to detect, and directly exports the result.As current subwindow is single pedestrian zone, then enters the next stage sorter and detects, and the complexity of sorter successively increases, and calculated amount is also increasing.Why cascade classifier can improve the speed of detection, in general input picture, most of zone is the single pedestrian of right and wrong zone all, and by these zones of simple sorter elimination, only the zone for a small amount of similar single pedestrian then needs more complicated sorter to detect.By this method, can detect single pedestrian in the present frame.
As described in the step 4 in the summary of the invention, need carry out modeling to single pedestrian's image that detection obtains.According to the position of human limb and the priori of structure, human body is divided into three parts, be respectively head, upper limbs and lower limb (as shown in Figure 6) are expressed as Head respectively, Upper and Lower, give different weights to various piece, actual conditions when blocking each other at pedestrian in the monitoring scene are given the highest weights to head, and lower limb are given minimum weights; Every single pedestrian's image that detection is obtained, calculate the residing position in the sensitizing range of every head, upper limbs and lower limb in pedestrian's image and shared area size, and the color histogram feature of various piece, this pedestrian is carried out modeling, obtain characterizing this pedestrian's model.Make T that (Location, Region Color) are feature from certain extracting section of pedestrian's human body, M (T
1, T
2, T
3, U, R, P) expression is to the model of pedestrian's foundation, M (T
1), M (T
2), M (T
3) represent from the head of manikin M these three features that part is extracted of upper limbs and lower limb, M (T respectively
i) (Location), M (T
i) (Region), M (T
i) (Color) (i=1,2,3) represent i of manikin M position, shared area size and the color histogram feature in the part respectively, M (U) is illustrated in this processing, whether model M is upgraded, M (R) represents the position of pedestrian's image of this model representative, and M (P) represents the historical position of this model.Single pedestrian's image that all detections are obtained carries out modeling, obtains model set S2.
As described in the step 5 in the summary of the invention, after detected single pedestrian is carried out modeling, adopt the method for template matches, the model among the S2 is mated.Each model M among the traversal S2
i, with each this model M do not upgraded among itself and the existing model set S1
jCompare, obtain similarity set (rejecting the model of all similarities) less than preset threshold.If the similarity set is not empty, suppose M
EFor among the S 1 with M
iThe model of similarity maximum is then with M
E(P) copy to M
i(P), with M
i(U) be set to upgrade, with M
iReplace the M among the S1
E, and from S2, delete M
iDo not have the model similar in the previous frame if the similarity set for empty, is illustrated in, think that the pedestrian of this model representative has just entered the sensitizing range, the number that enters the sensitizing range is added one to this model.
According to priori, the distance that same pedestrian moves in two continuous frames is no more than x pixel value, if M
NAnd M
ECenter of gravity (center with the model position characterizes) distance surpass x, then do not calculate M
NAnd M
ESimilarity.Make L
SimilarityRepresentation model M
NWith model M E
Similarity, then L
SimilarityCan be expressed as:
L
Similarity＝α*L
T(M
N(T
1)，M
E(T
1))+β*L
T(M
N(T
2)，M
E(T
2))+
γ*L
T(M
N(T
3)，M
E(T
3))+λ|R(M
N(R))-R(M
E(R))| (4)
α in the formula (4), beta, gamma, λ are the weights coefficient, the area of R (M (R)) representation model, L
TThe similarity of counterpart in three parts of two manikins is found the solution in expression, at M
NAnd M
EA same part when finding the solution similarity, L
T(M
N(T
k), M
E(T
k)) (k=1,2,3) can be expressed as:
L
T(M
N(T
k)，M
E(T
k))＝σ*|C(M
N(T
k)(Location))-C(M
E(T
k)(Location))|+
ω*|R(M
N(T
k)(Region))-R(M
E(T
k)(Region))+(5)
υ*P(M
N(T
k)(Color)，M
E(T
k)(Color))
σ in the formula (5), ω, υ are the weights coefficient, C (M
N(T
k) (Location)) representation model M
NIn K the part center of gravity, R (M
N(T
k) (Region)) representation model M
NIn K the part area, P (M
N(T
k) (Color), M
E(T
k) (Color)) expression employing histogram intersection method calculation model M
NAnd M
EIn the histogrammic distance of K part, suppose respectively two histograms of Q and D, then:
In the formula (6),
N is a pixel value number total in the histogram, n
kFor pixel value is the number of pixels of k, L represents the minimum value of the number of pixel values different in two histograms.
As described in the step 6 in the summary of the invention, behind all models in handling S2, check all models among the S1, if exist model not to be updated, represent that then the pedestrian of this model representative has left the sensitizing range.There is not the model M upgraded among the S1 any one
k, the number of leaving the sensitizing range is added one, the pedestrian's of this model representative movement locus is present in M
kIn P in.
Claims (5)
1. passenger flow statistical method based on adaptive template coupling, this method is to realize by a kind of passenger flow statistical system based on the adaptation module coupling, this passenger flow statistical system is made up of front-end and back-end, front end uses an industrial computer to connect image capture device, and monitor video is analyzed and added up; The rear end is a PC, monitor video and statistics that each front end sends is handled, and be saved in this locality; It is characterized in that: system front end is handled and is comprised the steps: that (1) front end industrial computer obtains frame of video from image capture device, supports to handle simultaneously multi-channel video; (2) according to the characteristics of passenger flow statistics, in the frame of video of obtaining, set the sensitizing range, only these regional data are handled; (3) use Adaboost algorithm to make up a plurality of Weak Classifiers based on the Haar feature, and these Weak Classifiers are connected from beginning to end, the output of a Weak Classifier is as the input of another sorter, obtain the strong classifier of a cascade, in the sensitizing range of video image, carry out single pedestrian's detection, obtain single pedestrian's image; (4) single pedestrian's image that detection is obtained carries out modeling, actual characteristic when blocking between the pedestrian, give different weights to the various piece of pedestrian's human body, be arranged in the position of frame of video, position and area size and the colouring information that the pedestrian partes corporis humani divides in conjunction with this pedestrian's image, for this pedestrian sets up model; (5) method of employing template matches is mated each model after the modeling and existing model, if it fails to match, thinks that then the pedestrian of this model representative has just entered the sensitizing range; If the match is successful, show that the pedestrian of this model representative has appeared in the sensitizing range, upgrade the overall positions of this model and the weight of each several part, make template matches have adaptive characteristic; (6) check existing model, if in from the current video frame sensitizing range, detecting the pedestrian dummy that also modeling obtains, do not have the model that is complementary with it, think that then the pedestrian of this model representative has left the sensitizing range, check the motion path of this model this moment, and carry out the statistics of passenger flow.
2. a kind of passenger flow statistical method according to claim 1 based on the adaptive template coupling, it is characterized in that: the sensitizing range described in the step (2) is the artificial rectangular area of setting of in the frame of video of being obtained in the step (1), and system front end is only handled the video data in this rectangular area.
3. a kind of passenger flow statistical method according to claim 1 based on the adaptive template coupling, it is characterized in that: the design of the strong classifier of the cascade described in the step (3) is by extracting single pedestrian's eigenwert and using the Adaboost algorithm to extract two parts of eigenwert design Weak Classifier and form, single pedestrian's feature of extracting has edge feature, symmetrical feature and central feature three major types, 12 group unique points altogether, and these unique points develop by the Haar feature; When using the Adaboost algorithm to extract eigenwert design Weak Classifier,, scan the central area computing center feature of the moving window of entire image earlier, then in the marginal portion of subwindow edge calculation feature according to single pedestrian's feature of image; After removing most of invalid rectangular characteristic according to these prioris, utilize the AdaBoost algorithm to pick out again to the best feature of single pedestrian's property distinguished, make up each Weak Classifier, when the false drop rate of Weak Classifier during less than setting threshold, stop to make up Weak Classifier, all Weak Classifiers that made up first places are linked to each other, obtain the strong classifier of a cascade, be used for the single pedestrian of sensitizing range detection.
4. a kind of passenger flow statistical method according to claim 1 based on the adaptive template coupling, it is characterized in that: step (4) is according to the position of human limb and the priori of structure, human body is divided into 3 parts, be respectively head, upper limbs and lower limb are given different weights to various piece, the actual conditions when blocking each other at pedestrian in the monitoring scene, give the highest weights to head, lower limb are given minimum weights; To detecting every the single pedestrian's image that obtains in the step (3), calculate the residing position in the sensitizing range of every head, upper limbs and lower limb in pedestrian's image and shared area size, and the color histogram feature of various piece, this pedestrian is carried out modeling, obtain characterizing this pedestrian's model.
5. a kind of passenger flow statistical method according to claim 1 based on the adaptive template coupling, it is characterized in that: after handling all models of being set up in the step (4), check existing model set T, if having certain is not initiate model, this model is not updated simultaneously, think that then the pedestrian of this model representative has left the sensitizing range, the motion path that check this model this moment carries out the renewal of the volume of the flow of passengers.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010105092177A CN101980245B (en) | 2010-10-11 | 2010-10-11 | Adaptive template matching-based passenger flow statistical method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010105092177A CN101980245B (en) | 2010-10-11 | 2010-10-11 | Adaptive template matching-based passenger flow statistical method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101980245A true CN101980245A (en) | 2011-02-23 |
| CN101980245B CN101980245B (en) | 2013-07-17 |
Family
ID=43600747
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010105092177A Expired - Fee Related CN101980245B (en) | 2010-10-11 | 2010-10-11 | Adaptive template matching-based passenger flow statistical method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101980245B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102842129A (en) * | 2012-05-24 | 2012-12-26 | 北京工业大学 | Non-standard template image matching method based on tree decomposition |
| CN103559537A (en) * | 2013-10-30 | 2014-02-05 | 南京邮电大学 | Template matching method based on error back propagation in out-of-order data streams |
| CN104166840A (en) * | 2014-07-22 | 2014-11-26 | 厦门亿联网络技术股份有限公司 | Focusing realization method based on video conference system |
| CN104766221A (en) * | 2015-03-25 | 2015-07-08 | 华迪计算机集团有限公司 | Method for selecting flow refining marketing parameters |
| CN104794443A (en) * | 2015-04-16 | 2015-07-22 | 杭州中威电子股份有限公司 | Self-help automatic teller machine realtime person number detection method based on image processing |
| CN107358149A (en) * | 2017-05-27 | 2017-11-17 | 深圳市深网视界科技有限公司 | A kind of human body attitude detection method and device |
| CN107679492A (en) * | 2017-09-29 | 2018-02-09 | 重庆市智权之路科技有限公司 | Behavior discriminant analysis method is carried out by using feature crawl function |
| CN108647631A (en) * | 2013-06-28 | 2018-10-12 | 日本电气株式会社 | Training data generates equipment, methods and procedures and crowd state identification equipment, methods and procedures |
| CN110838134A (en) * | 2019-10-10 | 2020-02-25 | 北京海益同展信息科技有限公司 | Target object statistical method and device, computer equipment and storage medium |
| CN111243339A (en) * | 2020-01-14 | 2020-06-05 | 武汉理工大学 | AIS system and method special for ferry with collision avoidance early warning function |
| CN114630238A (en) * | 2022-03-15 | 2022-06-14 | 广州康盛网络科技有限公司 | Stage loudspeaker box volume control method and device, electronic equipment and medium |
| CN115063378A (en) * | 2022-06-27 | 2022-09-16 | 中国平安财产保险股份有限公司 | Intelligent counting method, device, equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1503194A (en) * | 2002-11-26 | 2004-06-09 | 中国科学院计算技术研究所 | Status identification method by using body information matched human face information |
| CN101196991A (en) * | 2007-12-14 | 2008-06-11 | 同济大学 | Close passenger traffic counting and passenger walking velocity automatic detection method and system thereof |
| CN101216885A (en) * | 2008-01-04 | 2008-07-09 | 中山大学 | Passerby face detection and tracing algorithm based on video |
| CN101321269A (en) * | 2007-06-05 | 2008-12-10 | 同济大学 | Passenger flow volume detection method and system based on computer vision |
| EP2128818A1 (en) * | 2007-01-25 | 2009-12-02 | Shanghai Yaowei Industry Co, Ltd. | Method of moving target tracking and number accounting |
-
2010
- 2010-10-11 CN CN2010105092177A patent/CN101980245B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1503194A (en) * | 2002-11-26 | 2004-06-09 | 中国科学院计算技术研究所 | Status identification method by using body information matched human face information |
| EP2128818A1 (en) * | 2007-01-25 | 2009-12-02 | Shanghai Yaowei Industry Co, Ltd. | Method of moving target tracking and number accounting |
| CN101321269A (en) * | 2007-06-05 | 2008-12-10 | 同济大学 | Passenger flow volume detection method and system based on computer vision |
| CN101196991A (en) * | 2007-12-14 | 2008-06-11 | 同济大学 | Close passenger traffic counting and passenger walking velocity automatic detection method and system thereof |
| CN101216885A (en) * | 2008-01-04 | 2008-07-09 | 中山大学 | Passerby face detection and tracing algorithm based on video |
Non-Patent Citations (1)
| Title |
|---|
| 常好丽等: "基于单目视觉的运动行人检测与跟踪方法", 《交通运输工程学报》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102842129B (en) * | 2012-05-24 | 2014-12-31 | 北京工业大学 | Non-standard template image matching method based on tree decomposition |
| CN102842129A (en) * | 2012-05-24 | 2012-12-26 | 北京工业大学 | Non-standard template image matching method based on tree decomposition |
| US11132587B2 (en) | 2013-06-28 | 2021-09-28 | Nec Corporation | Training data generating device, method, and program, and crowd state recognition device, method, and program |
| CN108647631A (en) * | 2013-06-28 | 2018-10-12 | 日本电气株式会社 | Training data generates equipment, methods and procedures and crowd state identification equipment, methods and procedures |
| US11836586B2 (en) | 2013-06-28 | 2023-12-05 | Nec Corporation | Training data generating device, method, and program, and crowd state recognition device, method, and program |
| CN103559537A (en) * | 2013-10-30 | 2014-02-05 | 南京邮电大学 | Template matching method based on error back propagation in out-of-order data streams |
| CN103559537B (en) * | 2013-10-30 | 2016-03-23 | 南京邮电大学 | Based on the template matching method of error back propagation in a kind of out of order data stream |
| CN104166840A (en) * | 2014-07-22 | 2014-11-26 | 厦门亿联网络技术股份有限公司 | Focusing realization method based on video conference system |
| CN104766221A (en) * | 2015-03-25 | 2015-07-08 | 华迪计算机集团有限公司 | Method for selecting flow refining marketing parameters |
| CN104794443A (en) * | 2015-04-16 | 2015-07-22 | 杭州中威电子股份有限公司 | Self-help automatic teller machine realtime person number detection method based on image processing |
| CN104794443B (en) * | 2015-04-16 | 2018-09-21 | 杭州中威电子股份有限公司 | A kind of real-time number detection method of self-service automatic teller machine based on image procossing |
| CN107358149A (en) * | 2017-05-27 | 2017-11-17 | 深圳市深网视界科技有限公司 | A kind of human body attitude detection method and device |
| CN107679492A (en) * | 2017-09-29 | 2018-02-09 | 重庆市智权之路科技有限公司 | Behavior discriminant analysis method is carried out by using feature crawl function |
| CN110838134B (en) * | 2019-10-10 | 2020-09-29 | 北京海益同展信息科技有限公司 | Target object statistical method and device, computer equipment and storage medium |
| CN110838134A (en) * | 2019-10-10 | 2020-02-25 | 北京海益同展信息科技有限公司 | Target object statistical method and device, computer equipment and storage medium |
| CN111243339A (en) * | 2020-01-14 | 2020-06-05 | 武汉理工大学 | AIS system and method special for ferry with collision avoidance early warning function |
| CN114630238A (en) * | 2022-03-15 | 2022-06-14 | 广州康盛网络科技有限公司 | Stage loudspeaker box volume control method and device, electronic equipment and medium |
| CN115063378A (en) * | 2022-06-27 | 2022-09-16 | 中国平安财产保险股份有限公司 | Intelligent counting method, device, equipment and storage medium |
| CN115063378B (en) * | 2022-06-27 | 2023-12-05 | 中国平安财产保险股份有限公司 | Intelligent point counting method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101980245B (en) | 2013-07-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101980245B (en) | Adaptive template matching-based passenger flow statistical method | |
| Zhao et al. | Cloud shape classification system based on multi-channel cnn and improved fdm | |
| CN108830188B (en) | Vehicle detection method based on deep learning | |
| CN101859382B (en) | License plate detection and identification method based on maximum stable extremal region | |
| US10223597B2 (en) | Method and system for calculating passenger crowdedness degree | |
| US10867167B2 (en) | Collaborative deep network model method for pedestrian detection | |
| Zhou et al. | Robust vehicle detection in aerial images using bag-of-words and orientation aware scanning | |
| CN102332092B (en) | Flame detection method based on video analysis | |
| US8824742B2 (en) | Occupancy detection for managed lane enforcement based on localization and classification of windshield images | |
| CN102609686B (en) | Pedestrian detection method | |
| CN108921083B (en) | Illegal mobile vendor identification method based on deep learning target detection | |
| CN104091147B (en) | A kind of near-infrared eyes positioning and eye state identification method | |
| CN102043945B (en) | License plate character recognition method based on real-time vehicle tracking and binary index classification | |
| CN102609720B (en) | Pedestrian detection method based on position correction model | |
| CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
| CN102768726B (en) | Pedestrian detection method for preventing pedestrian collision | |
| CN106682586A (en) | Method for real-time lane line detection based on vision under complex lighting conditions | |
| CN111325146B (en) | Truck type and axle type identification method and system | |
| CN109583482A (en) | A kind of infrared human body target image identification method based on multiple features fusion Yu multicore transfer learning | |
| CN103761531A (en) | Sparse-coding license plate character recognition method based on shape and contour features | |
| CN101520841A (en) | Real-time and anti-interference method for positioning license plate in high-definition TV video | |
| CN106529532A (en) | License plate identification system based on integral feature channels and gray projection | |
| CN104615986A (en) | Method for utilizing multiple detectors to conduct pedestrian detection on video images of scene change | |
| CN104978567A (en) | Vehicle detection method based on scenario classification | |
| CN103632170A (en) | Pedestrian detection method and device based on characteristic combination |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20130717 Termination date: 20141011 |
|
| EXPY | Termination of patent right or utility model |