CN101872422B - People flow rate statistical method and system capable of precisely identifying targets - Google Patents
People flow rate statistical method and system capable of precisely identifying targets Download PDFInfo
- Publication number
- CN101872422B CN101872422B CN 201010114826 CN201010114826A CN101872422B CN 101872422 B CN101872422 B CN 101872422B CN 201010114826 CN201010114826 CN 201010114826 CN 201010114826 A CN201010114826 A CN 201010114826A CN 101872422 B CN101872422 B CN 101872422B
- Authority
- CN
- China
- Prior art keywords
- people
- sorter
- target trajectory
- rectangle
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000007619 statistical method Methods 0.000 title abstract description 4
- 238000001514 detection method Methods 0.000 claims abstract description 73
- 238000000034 method Methods 0.000 claims abstract description 47
- 238000012216 screening Methods 0.000 claims abstract description 26
- 238000007689 inspection Methods 0.000 claims description 4
- 239000002245 particle Substances 0.000 description 49
- 230000008569 process Effects 0.000 description 16
- 238000010586 diagram Methods 0.000 description 10
- 238000012549 training Methods 0.000 description 7
- 238000012544 monitoring process Methods 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 239000000284 extract Substances 0.000 description 4
- 230000006872 improvement Effects 0.000 description 4
- 230000015556 catabolic process Effects 0.000 description 2
- 238000007621 cluster analysis Methods 0.000 description 2
- 238000006731 degradation reaction Methods 0.000 description 2
- 238000009499 grossing Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000000903 blocking effect Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 239000012467 final product Substances 0.000 description 1
- 210000003141 lower extremity Anatomy 0.000 description 1
- 238000004377 microelectronic Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 230000007306 turnover Effects 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a people flow rate statistical method and a system capable of precisely identifying targets, wherein the method comprises the following steps: adopting a classifier to carry out human head rough detection on a current image, carrying out edge feature fine screening treatment on the rough detection result, and determining all human heads in the current image; tracking all the determined human heads, and forming a motion trajectory of human head targets; and carrying out people flow rate counting in the direction of the motion trajectory of the human head targets. Therefore, edge feature is further utilized for carrying out the fine screening on the rough detection result after the rough detection, thereby obtaining the real human head targets to a greater extent, improving the human head detection precision and further improving the accuracy of statistics of people flow rate.
Description
Technical field
The present invention relates to video monitoring and Flame Image Process and analysis technical field, but relate in particular to a kind of method and system of people flow rate statistical of precisely identifying targets.
Background technology
Along with the continuous progress of society, the range of application of video monitoring system is more and more wider.In the supermarket, the gateway in place such as market, gymnasium and station, airport often is equipped with rig camera, so that security personnel and supvr monitor the gateway in these places.On the other hand; The flow of the people of places such as supermarket, market, gymnasium and station, airport turnover has great significance for the operator in above-mentioned place or supvr; Wherein, Flow of the people is meant the number that flows by certain orientation, refers in particular among this paper by getting into/leave the number that both direction flows.
In the existing video monitoring, people flow rate statistical mainly is to check through monitor staff's manual work to realize.Reliable under the situation that the method for this complicate statistics flow of the people is short at monitoring period, flow of the people is sparse; But because the restriction of human eye biological nature; When monitoring period longer; When flow of the people is intensive, statistical accuracy will descend greatly, and the mode of complicate statistics needs the human cost of labor.Can realize the automatic statistics of flow of the people based on the people flow rate statistical method of video analysis, solve the variety of issue that complicate statistics brings.At present, the flow statistical method based on video analysis mainly contains three types:
One is based on the method for feature point tracking, and this method is at first followed the tracks of some motion characteristics points, and the track to unique point carries out cluster analysis then, thereby obtains flow of the people information; Need follow the tracks of some motion characteristics points based on the method for feature point tracking, the track to unique point carries out cluster analysis then, thereby obtains flow of the people information, and the shortcoming of this method is that unique point itself is difficult to stably follow the tracks of, and counting precision is relatively poor.
Two are based on the method that human body is cut apart and followed the tracks of, and this method at first need extract the moving target piece, the moving target piece are cut apart to obtain single human body target then, follow the tracks of each human body target at last and realize stream of people's quantitative statistics; Cut apart and the method for following the tracks of at first need extraction place moving target piece based on human body, the moving target piece is cut apart to obtain single human body target then, follow the tracks of the track that obtains each human body at last, thereby realize stream of people's quantitative statistics.The shortcoming of this method is that the accuracy that human body is cut apart is difficult to be guaranteed, and influences statistical precision when the human body existence is blocked.
Three are based on the method for the number of people or the detection and tracking of head shoulder, and this method detects the number of people or head shoulder in video, through stream of people's quantitative statistics is carried out in the tracking of the number of people or head shoulder.Method based on number of people detection and tracking is in video, to detect the number of people, carries out stream of people's quantitative statistics through the tracking to the number of people, when camera angle is suitable; The situation that blocking appears in the number of people is less, and the more preceding two kinds of method accuracys of method that therefore detect based on the number of people increase, and have company to propose the method based on number of people detection statistics number at present; For example Beijing Z-Star Microelectronics is in the method that the patent document of application number 200910076256.X is mentioned; At first extract sport foreground, adopt the sorter of haar features training dual serial in prospect, to search for the number of people of preliminary dimension then, realize that the number of people detects; Wherein, The haar characteristic is a kind of rectangular characteristic, through size that changes rectangle and shape and the half-tone information that array mode can be described target.This method only adopts sorter that target is carried out rough detection, and precision is not high, occurs the situation of number of people omission or many inspections easily, finally causes the not statistical uncertainty true problem of the number of people.
Summary of the invention
In view of this, but the present invention provides a kind of method and system of people flow rate statistical of precisely identifying targets, to solve the existing not statistical uncertainty true problem of people flow rate statistical scheme.
For this reason, the embodiment of the invention adopts following technical scheme:
But a kind of method of people flow rate statistical of precisely identifying targets comprises: adopt sorter that present image is carried out number of people rough detection, the rough detection result is carried out the edge feature fine screening handle, confirm each number of people in the present image; Each number of people to determining is followed the tracks of, and forms number of people target trajectory; Carry out the flow of the people counting according to number of people target trajectory direction.
Saidly the rough detection result is carried out the edge feature fine screening handle and to comprise: calculate said sorter and be judged as the rectangle inward flange characteristic of number of people target and the degree of fitting of the upside of ellipse arc that presets; If degree of fitting is greater than threshold value; Then this rectangle is confirmed as the number of people, otherwise this rectangle is removed from object listing.
Before the employing sorter carries out number of people rough detection to present image, also comprise: the surveyed area in the image is carried out scene calibration, thereby surveyed area is divided into plurality of sub-regions; Said sorter carries out number of people detection and in said plurality of sub-regions, carries out.
Saidly surveyed area in the image is carried out scene calibration comprise: select to demarcate frame; Calculate the scene depth variation factor; Calculate number of people target size variation range in the surveyed area; According to the number of people target size variation range surveyed area is divided into plurality of sub-regions.
After forming number of people target trajectory, according to number of people target trajectory direction, carry out also comprising: number of people target trajectory is carried out the smoothness analysis before the flow of the people counting.
Saidly number of people target trajectory is carried out the smoothness analysis comprise: confirm the smoothness of number of people target trajectory, judge whether said smoothness satisfies threshold value, if, keep this number of people target trajectory, otherwise, this number of people target trajectory abandoned.
Said sorter is the multicategory classification device of parallel connection.
Said employing sorter carries out number of people detection to image and comprises: the detection order that all kinds of sorters are set; Adopting each sorter that present image is carried out the number of people successively according to the detection order detects; Up to determining the number of people; Wherein, the multicategory classification device of said parallel connection is formed in parallel by at least two types of sorters.
The multicategory classification device of said parallel connection is formed in parallel by any two or more in dark hair generic classifier, light hair sorter, cap sorter and the expansion sorter.
But a kind of system of people flow rate statistical of precisely identifying targets; Comprise: number of people detection module; Be used to adopt sorter that present image is carried out number of people rough detection, and the number of people that Rough Inspection measures carried out the edge feature fine screening handle, confirm each number of people in the present image; Number of people target tracking module is used for each number of people of determining is followed the tracks of, and forms number of people target trajectory; The flow of the people counting module is used for carrying out the flow of the people counting in number of people target trajectory direction.
Also comprise: the scene calibration module is used for the surveyed area of image is carried out scene calibration, thereby surveyed area is divided into plurality of sub-regions.
Also comprise: number of people target trajectory analysis module, be used to calculate the smoothness of number of people target trajectory, judge whether said smoothness satisfies threshold value, if, keep this number of people target trajectory, otherwise, this number of people target trajectory abandoned.
Said sorter in the said number of people detection module is the multicategory classification device of parallel connection, and the multicategory classification device of said parallel connection is formed in parallel by any two or more in dark hair generic classifier, light hair sorter, cap sorter and the expansion sorter.
It is thus clear that the present invention utilizes edge feature that the rough detection result is carried out fine screening again, obtains real number of people target at last, make that detection is more accurate on the basis of sorter as number of people rough detection.Further, the present invention uses a plurality of sorter parallel connections, can detect multiclass number of people targets such as dark hair, light hair and shades of colour cap simultaneously, guarantees that statistics is more comprehensive.Further, the present invention also is provided with an expansion sorter, can be according to the application of particular surroundings, gather sample training, and detect the number of people of designated color or cap, such as the working cap in factory or warehouse etc.In addition, the present invention selects the size of detection window automatically through scene calibration before detection, makes the present invention can the various camera angle of self-adaptation, has widened range of application.And, can remove false target through smoothness analysis to number of people target trajectory, can further improve the detection accuracy rate.
Description of drawings
Fig. 1 is the method flow diagram of one embodiment of the invention people flow rate statistical;
Fig. 2 is the method flow diagram of another embodiment of the present invention people flow rate statistical;
Fig. 3 is a preferred embodiment scene calibration process flow diagram of the present invention;
Fig. 4 is a preferred embodiment number of people detection module structured flowchart of the present invention;
Fig. 5 is all kinds of sorter cascade of a preferred embodiment of the present invention assorting process synoptic diagram;
Fig. 6 is the process flow diagram of preferred embodiment particle filter tracking of the present invention;
Fig. 7 is preferred embodiment movement locus smoothness analysis process figure of the present invention;
Fig. 8 is the system architecture synoptic diagram of inventor's traffic statistics.
Embodiment
But the present invention proposes a kind of method of people flow rate statistical of precisely identifying targets, sees also Fig. 1, is one embodiment of the invention process flow diagram, comprising:
S101: adopt sorter that present image is carried out number of people rough detection, the rough detection result is carried out the edge feature fine screening handle, confirm each number of people in the present image;
S102: each number of people to determining is followed the tracks of, and forms number of people target trajectory;
S103: carry out the flow of the people counting according to number of people target trajectory direction.
Wherein, The rough detection result is carried out edge feature fine screening processing to be comprised: calculate said sorter and be judged as the rectangle inward flange characteristic of number of people target and the degree of fitting of the upside of ellipse arc that presets; If degree of fitting is greater than threshold value; Then this rectangle is confirmed as the number of people, otherwise this rectangle is removed from object listing.
It is thus clear that the present invention utilizes edge feature that the rough detection result is carried out fine screening again, obtains real number of people target at last, make that detection is more accurate on the basis of sorter as number of people rough detection
In order further to improve the accuracy of people flow rate statistical; On scheme basis shown in Figure 1, can further be optimized, comprise; The multicategory classification device of scene calibration, employing parallel connection carries out rough detection and number of people target trajectory is carried out smoothness analysis etc.; See also Fig. 2, be another embodiment of the present invention process flow diagram, comprising:
S201: scene calibration;
Particularly, scene calibration is meant carries out scene calibration to the surveyed area in the image, thereby surveyed area is divided into plurality of sub-regions.
S202: the number of people detects;
The number of people detects and further comprises parallelly connected sorter rough detection and two steps of edge feature fine screening, thereby confirms each number of people in the present image.
S203: number of people target following;
Through each number of people of determining is followed the tracks of, form number of people target trajectory.
S204: number of people target trajectory is carried out the smoothness analysis;
Particularly, number of people target trajectory is carried out the smoothness analysis comprise: confirm the smoothness of number of people target trajectory, judge whether said smoothness satisfies threshold value, if, keep this number of people target trajectory, otherwise, this number of people target trajectory abandoned.
S205: people flow rate statistical: flow of the people is counted through number of people target trajectory direction.
Need to prove, above-mentioned scene calibration, the number of people of parallelly connected sorter rough detection is carried out the edge feature fine screening, and can combine to use, also can use separately the improvement that number of people target trajectory is analyzed.
Carry out labor in the face of the optimum embodiment of the present invention who comprises all improvements down.
1, scene calibration
Because the video camera that is used for people flow rate statistical generally all is hard-wired, the scene variability is less, so the scene calibration module only need launch before first frame detects number of people target, and the result who all adopts first frame to demarcate when each frame detects number of people afterwards gets final product.If scene changes, then need launch scene calibration once more.
Under video camera situation without spin, the change in depth of scene can be approximated to be along the linear variation of image y coordinate, that is:
w(x，y)＝f×y+c (1)
Wherein, (x, y) expression center image coordinate is that (f is the scene depth coefficient for x, the width of number of people target boundary rectangle y), and c is a constant to w.The purpose of scene calibration is exactly to confirm the value of f and c through demarcating frame, thereby through type (1) is obtained the size of number of people target boundary rectangle in arbitrary coordinate place in the image.
Two unknown quantity f and the c of the present invention through selecting 4～6 to demarcate in the frame calculating formulas (1); Thereby obtain the change in depth coefficient of scene; In the coboundary and lower limb coordinate substitution formula (1) with the surveyed area boundary rectangle, obtain minimum people's area of bed w in the surveyed area then
MinWith maximum people's area of bed w
Max, last, according to the number of people change in size scope surveyed area is divided into plurality of sub-regions, corresponding one of each subregion changes less number of people range of size, and in ensuing number of people detection module, each subregion is with different size window search candidate rectangle.
Scene calibration step block diagram is as shown in Figure 3, comprising:
S301: select to demarcate frame;
S302: calculate the scene depth variation factor;
S303: calculate number of people target size variation range in the surveyed area;
S304: surveyed area is divided into plurality of sub-regions according to number of people target size variation range.
So far, scene calibration finishes.Next begin in each two field picture, to carry out detection, tracking and the counting of the number of people.
2, the number of people detects
The number of people among the present invention detects and is divided into parallelly connected sorter rough detection and two links of edge feature fine screening.
Through the good sorter of training in advance the inhuman head of major part is marked eliminating in the parallel connection sorter rough detection link; The inhuman head's mark of remaining number of people target and part flase drop behaviour head target; And then remove most of flase drop through edge feature fine screening link, keep real human head mark.Number of people detection module structured flowchart is as shown in Figure 4.
The present invention adopts the haar characteristic to train respectively based on the Adaboost algorithm to comprise the positive branch of the dark hair generic classifier of the front number of people and the back side number of people, dark hair sorter, dark hair back side branch sorter, light hair sorter, cap sorter and for adapting to the special a plurality of sorters such as expansion sorter that are provided with of specific environment.The array mode of a plurality of sorters is shown in Fig. 4 rough detection link: the positive branch of dark hair generic classifier and dark hair sorter, the synthetic tree structure of dark hair back side branch set of classifiers; Form parallelly connected with light hair sorter, cap sorter and expansion sorter then; The sorter testing result gets into people's head edge fine screening link, obtains real number of people target at last.
2.1, parallelly connected sorter rough detection link
Training aids needs to train with a large amount of positive samples and negative sample in advance, and the present invention adopts the haar characteristic of using in the detection of people's face to add Adaboost algorithm training recognizer.
The Haar characteristic is made up of the rectangle of two or three different sizes.The shape and the half-tone information of specific objective can be described through the size, array mode and the angle that change rectangle.The Adaboost algorithm is a kind of method that can some Weak Classifiers be combined into strong classifier.Each Weak Classifier selects one or several haar characteristic to come sample is classified, and several Weak Classifiers are through the synthetic one-level strong classifier of Adaboost algorithm groups.All kinds of sorters described in the present invention form by some grades of strong classifier cascades.
The present invention according to the number of people target size that the scene calibration module obtains, adopts the exhaustive mode seeker head to mark candidate rectangle in surveyed area.Candidate rectangle is input to respectively in dark hair generic classifier, light hair sorter, cap sorter and the expansion sorter classifies; If be classified as the number of people; Then this candidate rectangle is detected as the output of number of people target, continues to judge next candidate rectangle, otherwise; To select candidate rectangle to abandon, continue to judge next candidate rectangle.
In said process, a candidate rectangle is categorized as the strong classifiers at different levels that number of people target needs to pass through step by step cascade classifier by sorter, otherwise is classified as inhuman head's mark, and its process synoptic diagram is as shown in Figure 5.
In addition, in the above-mentioned sorter testing process, the preferential sorter of selecting can be adjusted according to practical application.The probability of dark hair is maximum in the general application scenarios, and the dark hair sorter of therefore preferential selection detects, and at special scenes, such as detecting the doorway, warehouse, the expansion sorter that can preferentially select the working cap sample training to obtain detects, to accelerate detection speed.
2.2, edge feature fine screening link
Through parallelly connected sorter rough detection link, most of non-number of people rectangle has been excluded, and only stays true number of people rectangle and is the rectangle of the number of people by the sorter flase drop.Edge feature fine screening link then can be removed most of flase drop rectangle through the edge feature that extracts in the rectangle, keeps real human head mark.
The present invention adopts oval first circular arc as the headform, and the edge feature fine screening is exactly to calculate by sorter to be judged as the rectangle inward flange characteristic of number of people target and the degree of fitting of oval first circular arc.If degree of fitting is greater than judgment threshold, then this rectangle is true number of people rectangle, otherwise is flase drop number of people rectangle, and this rectangle is removed from object listing.
3, the number of people is followed the tracks of
Need follow the tracks of after number of people target detection is come out, form target trajectory, to avoid same target repeat count.Target tracking module of the present invention adopts particle filter algorithm that number of people target is followed the tracks of.
The flow process of particle filter tracking is as shown in Figure 6, and detailed process is following:
Step 601: particle initialization;
When new detected number of people target does not have existing particle at once, a then newly-generated particle tracker, and with each particle position and size in the new detected object initialization tracker, and compose the weighted value of equating for each particle.
Step 602: particle resamples;
In tracing process; Particle " degradation phenomena " can occur through after weight is upgraded several times; Promptly the weight of the minority particle of approaching true number of people rectangle can become bigger; And becoming very little away from the weight of most of particle of number of people rectangle, a large amount of calculating can be wasted on the very little particle of these weights.In order to solve " degradation phenomena ", after upgrading, each particle weight should resample to particle.
It is exactly to keep and duplicate the bigger particle of weight that particle resamples, and rejects the less particle of weight, and the particle that makes the heavy particle of original cum rights be mapped as equal weight continues predicting tracing.When tracker was newly-generated, the weight of each particle equated in the tracker, therefore, need not resample again.
Step 603: the propagation of particle;
The propagation of particle also is the state transitions of particle, is meant the state renewal process in time of particle.Among the present invention, the state of particle is meant the position and the size of the target rectangle of particle representative.The propagation of particle adopts a kind of random motion process to realize that promptly the current state of particle adds that by Last status a random quantity obtains.Like this, each current particle is all being represented a possible position and the size of number of people target in present frame.
Step 604: according to observed reading new particle weight more;
Particle has just obtained possible position and the size of number of people target in present frame through circulation way, also need utilize the observed reading of present image to confirm that which particle most possibly is a number of people rectangle.Haar characteristic and the edge feature that extracts particle correspondence image rectangle among the present invention is as the observed reading weight of new particle more.The observed reading of particle is approaching more with the true number of people, and the rectangle that then this particle is corresponding possibly be number of people rectangle more, and the weight of particle increases; Otherwise the weight of particle reduces.
Step 605: upgrade target trajectory;
Particle is sorted by the weight size; Take out the maximum particle of weight, calculate the corresponding rectangle of the maximum particle of weight and detect the overlapping area that everyone head who obtains marks rectangle, overlapping area is maximum; And the number of people target greater than setting threshold promptly is the number of people of number of people target correspondence in present frame of this particle place tracker representative; Then use the target trajectory of the position renewal tracker of this number of people target, and replace the maximum particle of weight, get into next frame and follow the tracks of with this number of people target; If everyone head who detects in the maximum particle of weight and the present frame marks all not overlapping or overlapping area less than threshold value; Think that then the number of people target of this particle place tracker representative does not find the corresponding number of people in present frame; Then upgrade the target trajectory of tracker, and get into the next frame tracking with this particle position.If maximum particle N continuous (N＞2) frame of weight can not find corresponding number of people target, the number of people target and the disappearance of the tracker representative at this particle place then are described, reject this tracker.
Through above-mentioned five steps, the number of people target between frame and the frame just associates the movement locus that has formed number of people target.
4, smooth trajectory degree analysis module
In general, the motion of real human head's target is smoother, and the flase drop target then may present the mixed and disorderly motion in place, and therefore, the present invention removes flase drop through the smoothness analysis to target trajectory, further improves detection accuracy.
Target trajectory to tracking module generates is analyzed, and calculates the smoothing factor of target trajectory, if smoothing factor then keeps this track greater than the level and smooth threshold value of setting; Otherwise, reject this track.Smooth trajectory degree analysis module flow process is as shown in Figure 7, comprising:
S701: obtain target trajectory;
S702: the smoothness of confirming number of people target trajectory;
S703: judge whether smoothness satisfies the smoothness threshold value requirement of presetting, if, carry out S704, otherwise, S705 carried out;
S704: keep this target trajectory;
S705: abandon this target trajectory;
S706: export target movement locus.
5, flow of the people counting module
The present invention counts flow of the people through number of people target trajectory direction.The present invention judges in surveyed area whether the direction of this target trajectory is consistent with " people flows into " direction of setting, if consistent, then " entering number " counting adds one, otherwise " leaving number " counting adds one.Be " counting " with this target label after counting is accomplished, make track be in disarmed state, avoid same target repeat count.
So far, analyze and this five big step of people flow rate statistical, promptly accomplished comprehensive, accurate statistics flow of the people through scene calibration, number of people detection, number of people target following, number of people target trajectory.
Corresponding with said method, the present invention also provides a kind of system of people flow rate statistical, and this system can pass through software, hardware or software and hardware combining and realize.
With reference to figure 8, this system comprises:
Number of people detection module 801 is used to adopt sorter that present image is carried out number of people rough detection, and the number of people that Rough Inspection measures is carried out the edge feature fine screening handle, and confirms each number of people in the present image;
Number of people target tracking module 802, each number of people that is used for number of people detection module 801 is determined is followed the tracks of, and forms number of people target trajectory;
Flow of the people counting module 803 is used for carrying out the flow of the people counting in the number of people target trajectory direction that number of people target tracking module 802 is confirmed.
Preferably,
Number of people detection module 801 comprises rough detection submodule and fine screening submodule; The rough detection submodule is used to be provided with the detection order of all kinds of sorters; Adopting each sorter that present image is carried out the number of people successively according to the detection order detects; Up to determining the number of people, wherein, the multicategory classification device of said parallel connection is formed in parallel by at least two types of sorters.The multicategory classification device of the said parallel connection in the number of people detection module 801 is formed in parallel by any two or more in dark hair generic classifier, light hair sorter, cap sorter and the expansion sorter.The fine screening submodule is used for the detected number of people of multicategory classification device of parallel connection is carried out edge feature fine screening processing.
Preferably, this system also comprises:
Scene calibration module 804 is used for the surveyed area of image is carried out scene calibration, thereby surveyed area is divided into plurality of sub-regions.Wherein, the purpose of scene calibration module 804 is the depth coefficients that obtain scene, can calculate the size of the number of people target of each position in the image according to the scene depth coefficient, for number of people detection module 801 provides the detection size.At this moment, the size that number of people detection module 801 provides according to scene calibration module 804, seeker head's mark in the plurality of sub-regions of appointment.
Preferably, this system also comprises:
Number of people target trajectory analysis module 805 is used to calculate the smoothness of number of people target trajectory, judges whether said smoothness satisfies threshold value, if, keep this number of people target trajectory, otherwise, this number of people target trajectory abandoned.At this moment, flow of the people counting module 803 is on the basis of number of people target trajectory analysis module 805, adds up according to the number of people of movement locus direction.
The concrete realization of said system sees also method embodiment, does not give unnecessary details.
It is thus clear that the present invention utilizes edge feature that the rough detection result is carried out fine screening after rough detection again, obtains real number of people target more, improve the precision that the number of people detects, thereby improve the accuracy of people flow rate statistical.Preferably; A plurality of sorter parallel connections are used, can be detected multiclass number of people targets such as dark hair, light hair and shades of colour cap simultaneously, the present invention also is provided with an expansion sorter; Can be according to the application of particular surroundings; Gather sample training, detect the number of people of designated color or cap, such as the working cap in factory or warehouse etc.In addition, the present invention selects the size of detection window automatically through scene calibration before detection, makes the present invention can the various camera angle of self-adaptation, has widened range of application.And, can remove false target through smoothness analysis to number of people target trajectory, can improve the detection accuracy rate.
The above only is a preferred implementation of the present invention; Should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; Can also make some improvement and retouching, these improvement and retouching also should be regarded as protection scope of the present invention.
Claims (5)
1. but the method for the people flow rate statistical of a precisely identifying targets is characterized in that, comprising:
Adopt sorter that present image is carried out number of people rough detection, the rough detection result is carried out the edge feature fine screening handle, confirm each number of people in the present image; Saidly the rough detection result is carried out the edge feature fine screening handle and to comprise: calculate said sorter and be judged as the rectangle inward flange characteristic of number of people target and the degree of fitting of the upside of ellipse arc that presets; If degree of fitting is greater than threshold value; Then this rectangle is confirmed as the number of people, otherwise this rectangle is removed from object listing;
Each number of people to determining is followed the tracks of, and forms number of people target trajectory;
Calculate the smoothness of number of people target trajectory, judge whether said smoothness satisfies threshold value, if, keep this number of people target trajectory, otherwise, this number of people target trajectory abandoned;
Carry out the flow of the people counting in number of people target trajectory direction.
2. according to the said method of claim 1, it is characterized in that said sorter is the multicategory classification device of parallel connection.
3. according to the said method of claim 2, it is characterized in that the multicategory classification device of said parallel connection is formed in parallel by any two or more in dark hair generic classifier, light hair sorter and the cap sorter.
4. but the system of the people flow rate statistical of a precisely identifying targets is characterized in that, comprising:
Number of people detection module is used to adopt sorter that present image is carried out number of people rough detection, and the number of people that Rough Inspection measures is carried out the edge feature fine screening handle, and confirms each number of people in the present image; Saidly the rough detection result is carried out the edge feature fine screening handle and to comprise: calculate said sorter and be judged as the rectangle inward flange characteristic of number of people target and the degree of fitting of the upside of ellipse arc that presets; If degree of fitting is greater than threshold value; Then this rectangle is confirmed as the number of people, otherwise this rectangle is removed from object listing;
Number of people target tracking module is used for each number of people of determining is followed the tracks of, and forms number of people target trajectory;
Number of people target trajectory analysis module is used to calculate the smoothness of number of people target trajectory, judges whether said smoothness satisfies threshold value, if, keep this number of people target trajectory, otherwise, this number of people target trajectory abandoned;
The flow of the people counting module is used for carrying out the flow of the people counting in number of people target trajectory direction.
5. according to the said system of claim 4; It is characterized in that; Said sorter in the said number of people detection module is the multicategory classification device of parallel connection, and the multicategory classification device of said parallel connection is formed in parallel by any two or more in dark hair generic classifier, light hair sorter and the cap sorter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201010114826 CN101872422B (en) | 2010-02-10 | 2010-02-10 | People flow rate statistical method and system capable of precisely identifying targets |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201010114826 CN101872422B (en) | 2010-02-10 | 2010-02-10 | People flow rate statistical method and system capable of precisely identifying targets |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101872422A CN101872422A (en) | 2010-10-27 |
| CN101872422B true CN101872422B (en) | 2012-11-21 |
Family
ID=42997277
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201010114826 Active CN101872422B (en) | 2010-02-10 | 2010-02-10 | People flow rate statistical method and system capable of precisely identifying targets |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101872422B (en) |
Families Citing this family (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102004924A (en) * | 2010-11-03 | 2011-04-06 | 无锡中星微电子有限公司 | Human head detection system and method |
| CN102184409A (en) * | 2011-04-22 | 2011-09-14 | 北京文安科技发展有限公司 | Machine-vision-based passenger flow statistics method and system |
| US20130101159A1 (en) * | 2011-10-21 | 2013-04-25 | Qualcomm Incorporated | Image and video based pedestrian traffic estimation |
| CN102799935B (en) * | 2012-06-21 | 2015-03-04 | 武汉烽火众智数字技术有限责任公司 | Human flow counting method based on video analysis technology |
| CN102930334B (en) * | 2012-10-10 | 2013-08-14 | 北京凯森世纪科技发展有限公司 | Video recognition counter for body silhouette |
| CN102930618B (en) * | 2012-10-15 | 2015-07-15 | 北京理工大学 | Real-time people counting method and system based on ranging principle |
| CN105303191A (en) * | 2014-07-25 | 2016-02-03 | 中兴通讯股份有限公司 | Method and apparatus for counting pedestrians in foresight monitoring scene |
| CN104103077B (en) * | 2014-07-29 | 2017-04-19 | 浙江宇视科技有限公司 | Human head detecting method and human head detecting device |
| TWI537842B (en) * | 2014-09-30 | 2016-06-11 | 廣達電腦股份有限公司 | People counting system |
| CN106845318B (en) * | 2015-12-03 | 2019-06-21 | 杭州海康威视数字技术股份有限公司 | Passenger flow information acquisition method and device, passenger flow information processing method and processing device |
| CN106919889B (en) * | 2015-12-25 | 2021-04-13 | 株式会社日立制作所 | Method and device for detecting human head in video image |
| CN105654045B (en) * | 2015-12-29 | 2019-04-05 | 大连楼兰科技股份有限公司 | Method for identifying traffic controllers in autonomous driving technology |
| CN106570883B (en) * | 2016-10-25 | 2019-10-22 | 长安大学 | A People Counting Method Based on RGB-D Camera |
| CN108629230B (en) * | 2017-03-16 | 2021-02-12 | 杭州海康威视数字技术股份有限公司 | People counting method and device and elevator dispatching method and system |
| CN109389016B (en) * | 2017-08-10 | 2023-04-07 | 株式会社日立制作所 | Method and system for counting human heads |
| CN108875481B (en) * | 2017-08-31 | 2021-10-12 | 北京旷视科技有限公司 | Method, device, system and storage medium for pedestrian detection |
| CN109344690B (en) * | 2018-08-09 | 2022-09-23 | 上海青识智能科技有限公司 | People counting method based on depth camera |
| CN111339873B (en) * | 2020-02-18 | 2021-04-20 | 南京甄视智能科技有限公司 | Passenger flow statistical method and device, storage medium and computing equipment |
| CN111210555B (en) * | 2020-03-18 | 2024-08-06 | 三门峡云硕智能科技有限公司 | Intelligent warehouse management door |
| CN114581846B (en) * | 2022-03-03 | 2024-02-20 | 北京城建设计发展集团股份有限公司 | Real-time monitoring method and device for holographic passenger flow of subway station and computer equipment |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004295776A (en) * | 2003-03-28 | 2004-10-21 | Minolta Co Ltd | Image recognition device and image recognition program |
| KR100695136B1 (en) * | 2005-01-04 | 2007-03-14 | 삼성전자주식회사 | Image detection device and method |
| GB2431717A (en) * | 2005-10-31 | 2007-05-02 | Sony Uk Ltd | Scene analysis |
| CN101404086B (en) * | 2008-04-30 | 2012-05-09 | 浙江大学 | Target tracking method and device based on video |
| CN101447023B (en) * | 2008-12-23 | 2013-03-27 | 北京中星微电子有限公司 | Method and system for detecting human head |
| CN101477641A (en) * | 2009-01-07 | 2009-07-08 | 北京中星微电子有限公司 | Demographic method and system based on video monitoring |
| CN101464946B (en) * | 2009-01-08 | 2011-05-18 | 上海交通大学 | Detection method based on head identification and tracking characteristics |
-
2010
- 2010-02-10 CN CN 201010114826 patent/CN101872422B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN101872422A (en) | 2010-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101872422B (en) | People flow rate statistical method and system capable of precisely identifying targets | |
| CN101872431B (en) | People flow rate statistical method and system applicable to multi-angle application scenes | |
| CN101872414B (en) | People flow rate statistical method and system capable of removing false targets | |
| CN101877058B (en) | People flow rate statistical method and system | |
| US8798327B2 (en) | Method and system for people flow statistics | |
| CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
| US9418546B1 (en) | Traffic detection with multiple outputs depending on type of object detected | |
| CN103049787B (en) | A kind of demographic method based on head shoulder feature and system | |
| CN100361138C (en) | Method and system of real time detecting and continuous tracing human face in video frequency sequence | |
| CN109948582B (en) | A vehicle retrograde intelligent detection method based on tracking trajectory analysis | |
| CN102147869B (en) | Pedestrian detection method based on foreground analysis and pattern recognition | |
| CN102722698B (en) | Method and system for detecting and tracking multi-pose face | |
| CN108052859A (en) | A kind of anomaly detection method, system and device based on cluster Optical-flow Feature | |
| CN102521565A (en) | Garment identification method and system for low-resolution video | |
| CN103400391A (en) | Multiple-target tracking method and device based on improved random forest | |
| CN103324937A (en) | Method and device for labeling targets | |
| CN109508664A (en) | A kind of vegetable identification pricing method based on deep learning | |
| CN106886216A (en) | Robot automatic tracking method and system based on RGBD Face datections | |
| CN102842045A (en) | Pedestrian detection method based on combined features | |
| Lengvenis et al. | Application of computer vision systems for passenger counting in public transport | |
| CN109918971A (en) | Number detection method and device in monitor video | |
| CN106339657B (en) | Crop straw burning monitoring method based on monitor video, device | |
| KR102120812B1 (en) | Target recognition and classification system based on probability fusion of camera-radar and method thereof | |
| CN104346802A (en) | Method and device for monitoring off-job behaviors of personnel | |
| WO2018157056A1 (en) | Video data background tracking and subtraction with multiple layers of stationary foreground and regions |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| ASS | Succession or assignment of patent right |
Owner name: HANGZHOU HIKVISION DIGITAL TECHNOLOGY CO., LTD. Free format text: FORMER OWNER: HANGZHOU HAIKANG WEISHI SOFTWARE CO., LTD. Effective date: 20120905 |
|
| C41 | Transfer of patent application or patent right or utility model | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20120905 Address after: Hangzhou City, Zhejiang province 310012 Binjiang District East Road Haikang Science Park No. 700, No. 1 Applicant after: Hangzhou Hikvision Digital Technology Co., Ltd. Address before: Ma Cheng Road Hangzhou City, Zhejiang province 310012 No. 36 Applicant before: Hangzhou Haikang Weishi Software Co., Ltd. |
|
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |