CN104134078B - Automatic selection method for classifiers in people flow counting system - Google Patents
Automatic selection method for classifiers in people flow counting system Download PDFInfo
- Publication number
- CN104134078B CN104134078B CN201410349739.3A CN201410349739A CN104134078B CN 104134078 B CN104134078 B CN 104134078B CN 201410349739 A CN201410349739 A CN 201410349739A CN 104134078 B CN104134078 B CN 104134078B
- Authority
- CN
- China
- Prior art keywords
- target
- tracking
- grader
- detection
- tracking 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.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses an automatic selection method for classifiers in a people flow counting system, and belongs to the technical field of video monitoring and mode recognition. The automatic selection method comprises the following steps that: a plurality of imaging viewing angles are preset, and one classifier is respectively subjected to off-line training; during on-line studying, each classifier is respectively subjected to the following operations by aiming at a current input image: target detection is carried out, a current detected pedestrian target is correlated with an existing tracking target, uncorrelated tracking targets are tracked by adopting a tracking algorithm, and the uncorrelated detection targets are added into a tracking target queue; targets conforming to the conditions in the tracking target queue are counted; if the number of the targets counted by all of the classifiers does not exceed a threshold value, a next image is continuously processed; and the classifier with the most targets is selected to be used as the optimum classifier used in the people flow counting system. The method provided by the invention has the advantages that the optimum classifier can be automatically selected for the people flow counting system, and the problems of inaccuracy and inconvenience of manual setting of the classifiers are avoided.
Description
Technical field
The invention belongs to video monitoring and mode identification technology, more particularly, to a kind of people flow rate statistical system
The automatic selecting method of middle grader.
Background technology
In nineteen seventies, video monitoring system just has started to be occurred.With the development of science and technology, regard now
Frequency monitoring system generates very important impact to society, and CCTV camera almost spreads over streets and lanes.But it is actual
On, many monitor tasks that current monitoring system undertakes need for manual intervention.With the development of artificial intelligence technology, now
A kind of intelligent monitor system is risen.By computer vision technique the video information for collecting is carried out with mode identification technology
Analysis, extracts information useful in video and carries out associative operation, allows existing video monitoring system without artificial
In the case of intervention, some vision control tasks are automatically performed.
It is an important application in intelligent video monitoring system based on the people flow rate statistical technology of video.Flow of the people is always
All it is that market, supermarket, gymnasium and airport station etc. public place is managed and the indispensable significant data of decision-making.
For some industries, flow of the people can go out its business performance with direct reaction.In recent years, with intelligent video monitoring system
Rise, a research heat very with scientific meaning and commercial application value is become based on the people flow rate statistical technology of video
Point.People flow rate statistical data have very important meaning, and such as place such as airport station can utilize people flow rate statistical data
Estimate passenger flow information adjustment operation order of classes or grades at school, it is also possible to people flow rate statistical is carried out to certain gateway and then judges that the gateway sets
Whether reasonable put；Market supermarket can utilize people flow rate statistical data analysiss consumer buying habit, optimization StoreFront layout with
And assess the investment repayment of implemented marketing and promotion；And security personnel can prevent abnormal thing using people flow rate statistical data
The generation of part.
It is a kind of using relatively broad and statistical accuracy is higher in the existing people flow rate statistical method based on video
Method is that, based on the people flow rate statistical method of Statistical Learning Theory, the method needs first to extract the positive negative sample of target in scene,
Then choose a kind of feature description operator to describe target, feature is carried out to the positive negative sample for being gathered using machine learning algorithm
Learning classification, so as to obtain for the grader of target detection in people flow rate statistical.In people flow rate statistical, using above-mentioned study
Pedestrian target in the detection of classifier video scene for obtaining simultaneously is tracked to it, according to the movement locus of pedestrian target, system
The number of meter respective direction.
But the scene of people flow rate statistical application is needed due to different imagings, the angle of image of the CCTV camera installed
Degree is typically different.When using said method, in order to all obtain higher under different video imaging angles
Statistical accuracy, needs the grader for training each different imaging angle, then selects corresponding before people flow rate statistical algorithm application
The grader of angle carries out people flow rate statistical.Although improve the statistics essence under different imaging angles using the method for grader
Degree, but before people flow rate statistical system operation, the artificial corresponding grader of selection is needed, the adaptivity of system is reduced,
And because user may cause statistical accuracy to decline without the grader for correctly judging use.
The content of the invention
For the disadvantages described above or Improvement requirement of prior art, the present invention provides grader in a kind of people flow rate statistical system
Automatic selecting method, before people flow rate statistical system operation, the present invention selects optimum by the method for on-line study
Grader；Its object is to the manual setting grader for avoiding the people flow rate statistical algorithm based on Statistical Learning Theory from existing not
The problem of accuracy and inconvenience.
For achieving the above object, the present invention provides a kind of automatic selecting method of grader in people flow rate statistical system, bag
Include following steps：
(1) video source is obtained；
(2) two field picture in the video source is read as current image frame；
(3) all graders in people flow rate statistical system are collectively labeled as into unselected state, each grader correspondence is not
Same imaging angle；
(4) it is labeled as selecting a grader in the grader of non-selected state from all, and is marked as selecting
State；
(5) grader selected using the step (4) carries out target detection to the pedestrian in the current image frame；
(6) whether the tracking object queue for judging current selection sort device is empty, and if sky step (9) is then turned to, no
Step (7) is then turned to, wherein, the tracking object queue is the motion rail of the pedestrian target detected in the storage video source
The queue of mark；
(7) whether there is in the detection object queue of the grader selected described in judging and tracked in object queue with described
Tracking destination matches detection target, if it has, then the detection target and corresponding tracking target are collectively labeled as
With state, and with the positional information of corresponding tracking target described in the detection target update, by the corresponding tracking target
The frame number of times of appearance adds 1；Otherwise any operation is not carried out to tracking target, wherein, the detection object queue is described for storage
The queue of the pedestrian target detected in current image frame；
(8) whether each the tracking target in the tracking object queue of the grader selected described in judging is not
The state of matching, is that the tracking target not matched is tried to achieve using target tracking algorism in the position of the current image frame, and
With the positional information of the tracking target not matched described in the location updating tried to achieve；Otherwise any behaviour is not carried out to tracking target
Make；
(9) will be labeled as the detection target of matching status from it is described detection object queue in delete, if detection target with
Any tracking target is all mismatched in the tracking object queue of the grader for having selected, then added the detection target
Enter in the tracking object queue, and the frame number of times for being occurred plus 1；Otherwise any operation is not carried out to detecting target；
(10) if tracking target is located at the edge of the current image frame, it is deleted from the tracking object queue
Remove, otherwise do not carry out any operation to tracking target；
(11) tracking target to each in the tracking object queue of the grader for having selected respectively is carried out
Counting operation, obtains the number of the grader statistics for having selected；
(12) judge whether the grader for being also labeled as non-selected state, there is then execution step (4), otherwise execution step
(13)；
(13) judge whether the number that the most grader of statistical number of person counted exceedes statistical threshold, be then execution step
(14), otherwise execution step (2)；
(14) the most grader of the statistical number of person is selected as the most optimal sorting used in the people flow rate statistical system
Class device, to realize automatically selecting multiple graders in the people flow rate statistical system.
In general, by the contemplated above technical scheme of the present invention compared with prior art, with following beneficial effect
Really：
The present invention is a kind of method of on-line study, before operating in people flow rate statistical algorithm.Through to needing to select to divide
The people flow rate statistical scene video of class device is carried out after on-line study, and the present invention can be selected and be best suitable for this people flow rate statistical scene
Grader.After using the present invention, the fortune of the existing people flow rate statistical algorithm automatization based on Statistical Learning Theory can be allowed
OK, it is not necessary to artificial selection's optimum classifier, under it also avoid the statistical accuracy caused because artificially have selected wrong grader
Drop；
The present invention chooses classification score value highest and detects target as final detection target, the inspection that other are overlapped
Survey target to delete, reduce the possibility of false target appearance so that the target for detecting is substantially correct pedestrian target；
Present invention utilizes tracking target room and time information come to count target enter row constraint, for occurrence number
Less tracking target does not carry out counting operation, reduces the probability that false target is counted.
Description of the drawings
Fig. 1 is the flow chart of grader automatic selecting method in the present inventor's flow statistical system.
Fig. 2 (a) is a certain two field picture that the present invention is obtained from video source.
Fig. 2 (b) is the classification obtained by the positive and negative sample training extracted during the present invention adopts imaging angle for 45 degree of video
Device carries out the result of target detection to Fig. 2 (a).
Fig. 2 (c) is the result that the present invention carries out optimum classifier selection to certain video.
Specific embodiment
In order that the objects, technical solutions and advantages of the present invention become more apparent, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the present invention, and
It is not used in the restriction present invention.As long as additionally, technical characteristic involved in invention described below each embodiment
Not constituting conflict each other just can be mutually combined.
Fig. 1 show the flow chart of grader automatic selecting method in the present inventor's flow statistical system, including following step
Suddenly：
(1) video source is obtained.
(2) two field picture in the video source acquired in read step (1).Fig. 2 (a) show embodiment of the present invention video
A two field picture in source.
(3) unselected state will be collectively labeled as all graders in people flow rate statistical system.Preset many
Individual imaged viewing angle, difference one grader of off-line training.In present example, grader be based on imaging angle be 45 degree,
The linear SVM grader of 60 degree, three kinds obtained by 80 degree of positive and negative sample training different angles.
(4) from be labeled as in the grader of non-selected state select a grader, and labelling its be the state that selected.
(5) using the grader selected in step (4) to carrying out target detection as the pedestrian in previous frame image, concrete bag
Include following sub-step：
(5-1) from top to bottom, from left to right entered in the image of present frame successively with the detection window of different size size
Line slip is retrieved.In embodiments of the present invention, the size (resolution) of a two field picture is 352 × 288, the size bag of detection window
Include 22 × 22,26 × 26,30 × 30 and 34 × 34 4 kinds of sizes.
Extract the feature one of the target characteristic in current detection window, the target characteristic for being extracted and grader learning training
Cause.In embodiments of the present invention, the target characteristic of extraction is histograms of oriented gradients (Histogram of Oriented
Gradient, hereinafter referred to as HOG) feature.Histograms of oriented gradients is characterized in that by counting goal gradient intensity and directional spreding
To describe the face shaping feature of target, non-rigid (such as pedestrian) target can be well described, and scene can be resisted
In illumination variation.By score value of being classified accordingly in the current grader for selecting of target characteristic input of extraction, at this
In bright embodiment, the classification score value of target characteristic is typically distributed across that (0,4) in interval, the classification score value of non-targeted feature typically divides
Cloth (- 4,0) in interval.Classification score value score computing formula be：
Wherein [w1,w2,...,wq] it is the support vector of linear SVM grader；[x1,x2,...,xq] it is to search
HOG features in rope window；Q is the HOG intrinsic dimensionalities chosen；b*For the optimal classification that linear SVM grader is tried to achieve
Interval.
If the classification score value of gained more than classification thresholds (in embodiments of the present invention, the classification thresholds of selection are for 2.5),
Then assert there is target in the detection window, and the target for detecting be added in the detection object queue of the grader,
Wherein, detect that object queue is to deposit the queue of the pedestrian target detected in current image frame.In embodiments of the present invention, adopt
Realized detecting the data structure of object queue with the mode of chained list.Judging whether contain target in detection window using grader
When, it should the higher classification thresholds of setting, which decrease the possibility of false target appearance so that the target for detecting is basic
It is correct pedestrian target.
(5-2) detection window is moved to the next position of current image frame, and repeat step (5) until each size
Detection window all traveled through current image frame till.
(5-3) due to carrying out target detection using from top to bottom, from left to right traveling through by the way of picture frame, may cause big
Overlapping occurs in the detection target of amount.For such case, the embodiment of the present invention chooses the highest detection target conduct of classification score value
Final detection target, other detection targets for overlapping are deleted.It is positive and negative based on 45 degree of imagings that Fig. 2 (b) show selection
The pedestrian target that sample classification device is detected.
(6) whether the tracking object queue for judging current selection sort device is empty, if sky, then turns to step (9), no
Step (7) is then turned to, wherein, track the team that object queue is the movement locus for depositing the pedestrian target detected in video source
Row.
(7) each the tracking target in the tracking object queue of currently selected grader is traveled through, judges currently selected
Select in the detection object queue of grader with the presence or absence of the detection target with tracking destination matches.If it has, by the tracking mesh
Mark is collectively labeled as matching status with detection target, and with the positional information of the corresponding tracking target of the detection target update, by this
The frame number of times that tracking target occurs adds 1；Otherwise any operation is not carried out to the tracking target.
Specifically, if certain detection target and a tracking destination matches, then it is the distance between with the tracking target
Distance threshold is less than, and is less than in all detection targets of distance threshold and the tracking target with the tracking target range
Apparent colour information it is most like.In embodiments of the present invention, distance threshold value is 11 pixels；Apparent colour is similar
The calculating of degree adopts equation below：
Wherein, k represents tracking target；L represents detection target；By tracking target and the two of detection target place image-region
Dimension point set R={ It(a,b):a1≤a≤a2,b1≤b≤b2Be converted to one-dimensional vector X by row sequential storage, XkRepresent tracking
The one-dimensional vector of target, XlThe one-dimensional vector of detection target is represented, the average of one-dimensional vector X isN represents figure
As the number of pixel in region；XkiAnd XliRespectively vector XkAnd XiMiddle either element, vectorial XkRepresent tracking target, vector
XlRepresent detection target.
(8) each the tracking target in the tracking object queue of currently selected grader is traveled through, if certain tracking mesh
Mark is the state not matched, then be used for target tracking algorism and try to achieve the tracking target in the position of current image frame, is used in combination
The positional information of the location updating the tried to achieve tracking target；Otherwise, any operation is not carried out to the tracking target.Of the invention real
In applying example, selected track algorithm is mean shift algorithm.Mean shift algorithm is a kind of simple and effective track algorithm, should
Position of the algorithm after the extreme point of the constantly iteration searching object matching around target to be tracked is to try to achieve target following.
(9) each the detection target in the detection object queue of currently selected grader is traveled through, matching will be labeled as
The detection target of state is deleted from detection object queue.If certain detects the tracking target of target and currently selected grader
Any tracking target is all mismatched in queue, then the detection target is added in tracking object queue, and occurred
Frame number of times adds 1；Otherwise, any operation is not carried out to the detection target.
(10) each the tracking target in the tracking object queue of current selection sort device is traveled through, if the tracking target
Positioned at the edge of current image frame, then it is deleted from tracking object queue；Otherwise any behaviour is not carried out to the tracking target
Make.In embodiments of the present invention, if the center of tracking target is less than 17 pixels with the distance on current image frame one side
Point (half of maximum detection window size 34 × 34 i.e. in the embodiment of the present invention), then it is assumed that the tracking target has been positioned at currently
The edge of picture frame.
(11) each the tracking target in the tracking object queue of currently selected grader is traveled through, it is carried out respectively
Counting operation, specifically includes following sub-step：
(11-1) judge whether the tracking target is labeled as count status, if it is, not carrying out to the tracking target
Any operation, otherwise turns to step (11-2)；
(11-2) judge whether the tracking target is more than in the position of previous frame image with the position difference of current frame image
Potential difference threshold value, in embodiments of the present invention, potential difference threshold value value is 3 pixels.If it is, step (11-3) is proceeded to, it is no
Then any operation is not carried out to the tracking target；
(11-3) judge whether the frame number of times that the tracking target occurs is more than occurrence number threshold value, in the embodiment of the present invention
In, occurrence number threshold value value is 3.If it is, adding 1 by people's numerical value that currently selected grader is counted, otherwise do not carry out
Any operation.
The advantage of step (11-2) and (11-3) is that the room and time information that make use of tracking target is come to counting target
Enter row constraint, the tracking target less for occurrence number does not carry out counting operation, reduces the probability that false target is counted.
(12) grader for being also labeled as non-selected state is judged whether, having then turn to step (4), otherwise turns to step
(13)。
(13) whether the number for judging the most grader statistics of statistical number of person exceedes statistical threshold, in the embodiment of the present invention
In, statistical threshold value is 100, if being less than the statistical threshold, turns to step (2), otherwise turns to step (14).
(14) the most grader of statistical number of person is selected as the optimum classifier used in follow-up people flow rate statistical system,
To realize automatically selecting multiple graders in people flow rate statistical system.Fig. 2 (c) show the present invention and certain video is carried out point
The result that class device is selected, textual representation is directed to this video in figure, and 45 degree of graders count 100 people, and 60 degree of graders are counted
67 people, 80 degree of graders count 50 people, the final principle that optimum classifier is up to according to the number for counting, and have selected 45
Degree grader.
As it will be easily appreciated by one skilled in the art that the foregoing is only presently preferred embodiments of the present invention, not to
The present invention, all any modification, equivalent and improvement made within the spirit and principles in the present invention etc. are limited, all should be included
In protection scope of the present invention.
Claims (6)
1. in a kind of people flow rate statistical system grader automatic selecting method, it is characterised in that comprise the following steps：
(1) video source is obtained；
(2) two field picture in the video source is read as current image frame；
(3) all graders in people flow rate statistical system are collectively labeled as into unselected state, each grader correspondence it is different into
Image angle degree；
(4) it is labeled as selecting a grader in the grader of non-selected state from all, and is marked as selecting state；
(5) grader selected using the step (4) carries out target detection to the pedestrian in the current image frame；
(6) whether the tracking object queue for judging current selection sort device is empty, and if empty step (9) is then turned to, and is otherwise turned
To step (7), wherein, the tracking object queue is the movement locus of the pedestrian target detected in the storage video source
Queue；
(7) whether there is in the detection object queue for judging the grader for having selected with it is described track in object queue with
The detection target of track destination matches, if it has, then being collectively labeled as matching shape with corresponding tracking target by the detection target
State, and with the positional information of corresponding tracking target described in the detection target update, the corresponding tracking target is occurred
Frame number of times add 1；Otherwise any operation is not carried out to tracking target, wherein, the detection object queue is described current for storage
The queue of the pedestrian target detected in picture frame；
(8) whether each the tracking target in the tracking object queue of the grader selected described in judging is not match
State, be that the tracking target not matched is tried to achieve using target tracking algorism in the position of the current image frame, and use institute
State the positional information of the tracking target not matched described in the location updating tried to achieve；Otherwise any operation is not carried out to tracking target；
(9) will be labeled as the detection target of matching status from it is described detection object queue in delete, if detection target with it is described
Any tracking target is all mismatched in the tracking object queue of the grader for having selected, then be added to the detection target
In the tracking object queue, and the frame number of times for being occurred plus 1；Otherwise any operation is not carried out to detecting target；
(10) if tracking target is located at the edge of the current image frame, it is deleted from the tracking object queue,
Otherwise any operation is not carried out to tracking target；
(11) track target to each in the tracking object queue of the grader for having selected respectively to count
Operation, obtains the number of the grader statistics for having selected；
(12) judge whether the grader for being also labeled as non-selected state, there is then execution step (4), otherwise execution step
(13)；
(13) judge whether the number that the most grader of statistical number of person counted exceedes statistical threshold, be then execution step
(14), otherwise execution step (2)；
(14) the most grader of the statistical number of person is selected as the optimum classifier used in the people flow rate statistical system,
To realize automatically selecting multiple graders in the people flow rate statistical system.
2. the method for claim 1, it is characterised in that the step (5) specifically includes following sub-step：
(5-1) from top to bottom, from left to right carried out in the present image frame in successively with the detection window of different size size
Slide and retrieve, extract the target characteristic in current detection window, the classification score value score of the target characteristic is calculated, if institute
The classification score value score be more than classification thresholds, then assert and exist in the detection window detection target, and by the inspection
Survey target to be added in the detection object queue of the grader for having selected, wherein, the calculating of the score value score that classifies is public
Formula is：
Wherein, [w1,w2,...,wq] represent linear SVM grader support vector；[x1,x2,...,xq] represent institute
State the histograms of oriented gradients feature in detection window；Q represents the histograms of oriented gradients intrinsic dimensionality of selection；b*Represent described
The optimal classification interval that linear SVM grader is tried to achieve；
(5-2) detection window is moved to the next position of the current image frame, and repeating said steps (5-1) until every
Till the detection window of individual size has all traveled through the current image frame；
(5-3) the classification score value score highests are chosen and detects that target, as final detection target, other is overlapped
Detection target delete.
3. method as claimed in claim 1 or 2, it is characterised in that judge in the step (7) detection target whether with tracking
The method of destination matches is：If detection target and a tracking destination matches, then it is between the tracking target
Distance is less than distance threshold, and be in all detection targets with the tracking target range less than the distance threshold with
The apparent colour information of the tracking target is most like, wherein, the apparent colour similarity is counted using equation below
Calculate：
Wherein, k represents tracking target；L represents detection target；By tracking target and the two-dimensional points of detection target place image-region
Collection R={ It(a,b):a1≤a≤a2,b1≤b≤b2Be converted to one-dimensional vector X by row sequential storage, XkRepresent tracking target
One-dimensional vector, XlThe one-dimensional vector of detection target is represented, the average of one-dimensional vector X isN represents institute
State the number of pixel in image-region；XkiAnd XliRespectively vector XkAnd XlMiddle either element.
4. method as claimed in claim 1 or 2, it is characterised in that the target tracking algorism used in the step (8)
For mean shift algorithm.
5. method as claimed in claim 2, it is characterised in that the step (10) if in track center and the institute of target
State the half of the distance less than maximum detection window size on current image frame one side, then it is assumed that the tracking target has been positioned at
The edge of the current image frame.
6. method as claimed in claim 1 or 2, it is characterised in that the step (11) specifically includes following sub-step：
(11-1) judge to track whether target is labeled as count status, if it is, not carrying out the tracking target any
Operation, otherwise execution step (11-2)；
(11-2) judge to track whether target is more than in the position of previous frame image with the difference of the position of the current frame image
Potential difference threshold value, if it is execution step (11-3), does not otherwise carry out any operation to the tracking target；
(11-3) whether judge to track frame number of times that target occurs more than occurrence number threshold value, if it is selected described
People's numerical value of grader statistics adds 1, does not otherwise carry out any operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410349739.3A CN104134078B (en) | 2014-07-22 | 2014-07-22 | Automatic selection method for classifiers in people flow counting system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410349739.3A CN104134078B (en) | 2014-07-22 | 2014-07-22 | Automatic selection method for classifiers in people flow counting system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104134078A CN104134078A (en) | 2014-11-05 |
| CN104134078B true CN104134078B (en) | 2017-04-26 |
Family
ID=51806752
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410349739.3A Expired - Fee Related CN104134078B (en) | 2014-07-22 | 2014-07-22 | Automatic selection method for classifiers in people flow counting system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104134078B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105224922B (en) * | 2015-09-18 | 2018-10-16 | 桂林远望智能通信科技有限公司 | A kind of people's vehicle classification video streaming processing system and method |
| CN105631418B (en) * | 2015-12-24 | 2020-02-18 | 浙江宇视科技有限公司 | People counting method and device |
| CN107491715A (en) * | 2016-06-13 | 2017-12-19 | 北京文安智能技术股份有限公司 | A kind of subway carriage passenger flow statistical method, apparatus and system based on video analysis |
| CN106485217B (en) * | 2016-09-30 | 2019-10-25 | 四川世纪云道科技有限公司 | A kind of method and system of identification visit scenic spot stream of people saturation degree |
| CN107153819A (en) * | 2017-05-05 | 2017-09-12 | 中国科学院上海高等研究院 | A kind of queue length automatic testing method and queue length control method |
| US10678848B2 (en) | 2018-02-12 | 2020-06-09 | Wipro Limited | Method and a system for recognition of data in one or more images |
| CN108921072B (en) * | 2018-06-25 | 2021-10-15 | 苏州欧普照明有限公司 | People flow statistical method, device and system based on visual sensor |
| CN109784260A (en) * | 2019-01-08 | 2019-05-21 | 深圳英飞拓科技股份有限公司 | A kind of zone flow real-time statistical method and system based on video structural |
| CN110532883B (en) * | 2019-07-30 | 2023-09-01 | 平安科技（深圳）有限公司 | Improvement of on-line tracking algorithm by off-line tracking algorithm |
| CN111310652B (en) * | 2020-02-13 | 2023-07-25 | 平安科技（深圳）有限公司 | Flow identification method and system based on classification regression double-domain model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102289817A (en) * | 2011-07-19 | 2011-12-21 | 中国科学院自动化研究所 | pedestrian counting method based on group context |
| CN102663409A (en) * | 2012-02-28 | 2012-09-12 | 西安电子科技大学 | Pedestrian tracking method based on HOG-LBP |
| CN103077403A (en) * | 2012-12-30 | 2013-05-01 | 信帧电子技术（北京）有限公司 | Pedestrian counting method and device |
| CN103902968A (en) * | 2014-02-26 | 2014-07-02 | 中国人民解放军国防科学技术大学 | Pedestrian detection model training method based on AdaBoost classifier |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008008505A2 (en) * | 2006-07-14 | 2008-01-17 | Objectvideo, Inc. | Video analytics for retail business process monitoring |
-
2014
- 2014-07-22 CN CN201410349739.3A patent/CN104134078B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102289817A (en) * | 2011-07-19 | 2011-12-21 | 中国科学院自动化研究所 | pedestrian counting method based on group context |
| CN102663409A (en) * | 2012-02-28 | 2012-09-12 | 西安电子科技大学 | Pedestrian tracking method based on HOG-LBP |
| CN103077403A (en) * | 2012-12-30 | 2013-05-01 | 信帧电子技术（北京）有限公司 | Pedestrian counting method and device |
| CN103902968A (en) * | 2014-02-26 | 2014-07-02 | 中国人民解放军国防科学技术大学 | Pedestrian detection model training method based on AdaBoost classifier |
Non-Patent Citations (3)
| Title |
|---|
| "Counting moving persons in crowded scenes";Donatello Conte et al.;《Machine Vision and Applications》;20131231;第2013年卷(第24期);第1029-1042页 * |
| "Online Multiperson Tracking-by-Detection from a Single, Uncalibrated Camera";Michael D. Breitenstein et al.;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20110930;第33卷(第9期);第1820-1833页 * |
| "多通道Reichardt运动目标检测算法";孙彬 等;《华中科技大学学报》;20101031;第38卷(第10期);第57-60页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104134078A (en) | 2014-11-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104134078B (en) | Automatic selection method for classifiers in people flow counting system | |
| CN103914702B (en) | System and method for improving the object detection performance in video | |
| CN109948582B (en) | Intelligent vehicle reverse running detection method based on tracking trajectory analysis | |
| US9129524B2 (en) | Method of determining parking lot occupancy from digital camera images | |
| EP2659456B1 (en) | Scene activity analysis using statistical and semantic feature learnt from object trajectory data | |
| KR101592889B1 (en) | Object matching for tracking, indexing, and search | |
| EP3002710A1 (en) | System and method for object re-identification | |
| CN109829382B (en) | Abnormal target early warning tracking system and method based on intelligent behavior characteristic analysis | |
| CN105512640A (en) | Method for acquiring people flow on the basis of video sequence | |
| Derpanis et al. | Classification of traffic video based on a spatiotemporal orientation analysis | |
| KR101697161B1 (en) | Device and method for tracking pedestrian in thermal image using an online random fern learning | |
| Romdhane et al. | An improved traffic signs recognition and tracking method for driver assistance system | |
| CN104615986A (en) | Method for utilizing multiple detectors to conduct pedestrian detection on video images of scene change | |
| Juranek et al. | Real-time pose estimation piggybacked on object detection | |
| CN104978567A (en) | Vehicle detection method based on scenario classification | |
| Liu et al. | Multi-type road marking recognition using adaboost detection and extreme learning machine classification | |
| Parameswaran et al. | Design and validation of a system for people queue statistics estimation | |
| Zhang et al. | A Multiple Instance Learning and Relevance Feedback Framework for Retrieving Abnormal Incidents in Surveillance Videos. | |
| Gabb et al. | Efficient monocular vehicle orientation estimation using a tree-based classifier | |
| Patel et al. | Vehicle tracking and monitoring in surveillance video | |
| CA3012927A1 (en) | Counting objects in images based on approximate locations | |
| Chen et al. | Context-aware lane marking detection on urban roads | |
| KR20170104756A (en) | Local size specific vehicle classifying method and vehicle detection method using the classifying method | |
| Rouf et al. | Real-time Vehicle Detection, Tracking and Counting System Based on YOLOv7 | |
| Thokala et al. | Accident Prevention System using Machine Learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170426 Termination date: 20190722 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |