CN101996400A - Method and device for updating target detector - Google Patents
Method and device for updating target detector Download PDFInfo
- Publication number
- CN101996400A CN101996400A CN200910166494XA CN200910166494A CN101996400A CN 101996400 A CN101996400 A CN 101996400A CN 200910166494X A CN200910166494X A CN 200910166494XA CN 200910166494 A CN200910166494 A CN 200910166494A CN 101996400 A CN101996400 A CN 101996400A
- Authority
- CN
- China
- Prior art keywords
- sub
- classifier
- object detector
- confidence level
- pond
- 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 a method and device for updating a target detector. The target detector is composed of a plurality of sub-classifiers and is used for detecting a target in a video. The method comprises the following steps: judging credibility: judging whether the credibility of the target detector meets a predetermined updating condition; and updating: if the credibility meets the predetermined updating condition, replacing at least one sub-classifier with the worst classification capability in the plurality of the sub-classifiers with a new sub-classifier. In the invention, at least one sub-classifier with the worst classification capability in the plurality of the sub-classifiers is updated according to the credibility of the target detector during the detection process so that the target detector can adapt to status change of the target in the video, thus realizing robust detection and tracking of the target.
Description
Technical field
Present invention relates in general to target tracking domain, more specifically, the present invention relates to a kind of method and apparatus that upgrades object detector.
Background technology
Target following is a major issue of computer vision field, and it is the basis of various higher layer applications such as behavioural analysis, abnormality detection, has great significance in video monitoring system.Tracking comes down to find the dbjective state problem of (being used to characterize position, size, color and other parameter of target) from the image sequence that observes.Though the someone carries out the research of target following aspect very early, the target following of robust is very challenging problem in this area always.
For robust ground tracing object in video, people have introduced various features, such as gradient orientation histogram (Histogram of Oriented Gradient, HOG), Haar wavelet character, local binary pattern (Local Binary Pattern, LBP) or the like.
Yet in tracing process, the state of target can change, and these states that changed are difficult situations that detect or can't detect at all of object detector of utilizing off-line training to obtain.The state variation of target has a lot of situations, wherein the variation of the variation of illumination, target pose and block (comprise target by other object block, target occlusion other object and target blocked by background) be most important three kinds of situations.
Be tracked as example with the pedestrian, the difficulty that exists in the target following of prior art is described.Fig. 1 illustrates the video segment that the pedestrian turns to the side.As shown in Figure 1, when pedestrian in the video transferred the front to by the side or transfers the side to by the front, its posture change was bigger.In addition, when showing the pedestrian side in video, along with pedestrian's walking, the posture change of its shank is also bigger.In this case, the pedestrian detector who obtains with off-line training detects the pedestrian with being difficult to robust.Fig. 2 illustrates the video segment that the pedestrian squats down.As shown in Figure 2, when the pedestrian squatted down from uprightly becoming, whole people's posture change was very big.In this case, the pedestrian detector who obtains with off-line training can't detect the pedestrian after squatting down at all.
Be example with the vehicle tracking again, the difficulty that exists in the target following of prior art is described.When vehicle such as turn, during doubling, bigger variation takes place in the posture of vehicle.In addition, when illumination condition changed, corresponding variation can appear in the color of vehicle.In addition, the vehicle on the road may often be blocked or block other vehicle by other vehicle, is perhaps blocked by the backgrounds such as trees of roadblock, road both sides.In these cases, the wagon detector that obtains with off-line training detects target vehicle with being difficult to robust.
Summary of the invention
At above and other problem, the present invention proposes a kind of method and apparatus that upgrades object detector.Method and apparatus according to the embodiment of the invention makes object detector can adapt to the state variation of target in video.
According to an aspect of the present invention, provide a kind of method of upgrading object detector.Described object detector is made up of a plurality of sub-classifiers, is used to detect objects in video.Described method comprises: the confidence level determining step, judge whether the confidence level of described object detector satisfies the scheduled update condition; And step of updating, if described confidence level satisfies the scheduled update condition, then replace the minimum sub-classifier of at least one classification capacity in described a plurality of sub-classifier with new sub-classifier.
According to a further aspect in the invention, provide a kind of device that upgrades object detector.Described object detector is made up of a plurality of sub-classifiers, is used to detect objects in video.Described device comprises: upgrade judging unit, described renewal judging unit comprises the confidence level judging unit, and described confidence level judging unit is configured to judge whether the confidence level of described object detector satisfies the scheduled update condition; And the renewal performance element, be configured to:, then replace the minimum sub-classifier of at least one classification capacity in described a plurality of sub-classifier with new sub-classifier if described confidence level satisfies the scheduled update condition.
According to the method and apparatus of renewal object detector of the present invention by in online testing process, upgrading the minimum sub-classifier of at least one classification capacity in the object detector according to the confidence level of object detector, make the state variation of the target of object detector in can adaptive video, thus robust ground detection and tracking target.
According to a further aspect in the invention, also provide a kind of storage medium.Described storage medium comprises machine-readable program code, and when carrying out described program code on messaging device, described program code makes described messaging device carry out the method according to renewal object detector of the present invention.
According to a further aspect in the invention, also provide a kind of program product.Described program product comprises the executable instruction of machine, and when carrying out described instruction on messaging device, described instruction makes described messaging device carry out the method according to renewal object detector of the present invention.
Description of drawings
Above and other purpose of the present invention, feature and advantage will be by being better understood with reference to hereinafter given in conjunction with the accompanying drawings description.In institute's drawings attached, same or analogous Reference numeral is represented identical or similar parts.In described accompanying drawing:
Fig. 1 illustrates the video segment that the pedestrian turns to the side;
Fig. 2 illustrates the video segment that the pedestrian squats down;
Fig. 3 is the exemplary diagram of position of preceding 20 pairing HOG pieces of sub-classifier of employed object detector in the embodiments of the invention;
Fig. 4 is the process flow diagram that the method for upgrading object detector according to an embodiment of the invention is shown;
Fig. 5 illustrates the process flow diagram of step of updating according to an embodiment of the invention;
Fig. 6 is the process flow diagram that step of updating according to another embodiment of the present invention is shown;
Schematically illustrated 10 the front samples of Fig. 7 A;
Fig. 7 B is shown schematically in the diagrammatic sketch of 3 sub-classifiers removing in the step of updating according to the method for the renewal object detector of the embodiment of the invention from object detector;
Fig. 7 C is shown schematically in the diagrammatic sketch of 3 new sub-classifiers selecting in the step of updating according to the method for the renewal object detector of the embodiment of the invention;
Fig. 8 is the process flow diagram that the method for renewal object detector according to another embodiment of the present invention is shown;
Fig. 9 is a schematic block diagram of upgrading the device of object detector according to an embodiment of the invention;
Figure 10 is the schematic block diagram of the device of renewal object detector according to another embodiment of the present invention;
Figure 11 shows the schematic block diagram that can be used for implementing according to the computing machine of the method and apparatus of the embodiment of the invention.
Embodiment
Embodiments of the invention are described with reference to the accompanying drawings.Element of describing in an accompanying drawing of the present invention or a kind of embodiment and feature can combine with element and the feature shown in one or more other accompanying drawing or the embodiment.Should be noted that for purpose clearly, omitted the parts that have nothing to do with the present invention, those of ordinary skills are known and the expression and the description of processing in accompanying drawing and the explanation.
Object detector
There is the multiple object detector that constitutes by a plurality of sub-classifiers in the prior art, for example, a kind of such object detector is disclosed in people's such as Marco Pedersoli article " Boosting Histograms of Oriented Gradientsfor Human Detection " (http://lear.inrialpes.fr/people/triggs/pubs/Dalal-cvpr05.pdf, last visit on August 11st, 2009).Convenient in order hereinafter to describe, object detector is once at first briefly described.
As example, (Histogramof Oriented Gradient, HOG) feature is come the training objective detecting device in conjunction with gradient orientation histogram to use enhancing (Boosting) method.Using the benefit of Boosting method is to access a more continuous output, makes the renewal of tracking and object detector be more prone to realize.Equally, can carry out on robust ground in order to make renewal, each statistical value is as a Weak Classifier (corresponding to sub-classifier) in the application HOG feature.The main theoretical basis of doing like this is that the gradient direction of the last the first has occupied leading position in each HOG piece.Therefore, only use a statistical value just to represent the information in this piece to a great extent.Like this, can in an image, obtain a plurality of Weak Classifiers (for hereinafter describing conveniently, be designated as M Weak Classifier, M is a natural number).The target image that can use a plurality of manual demarcation is as positive test sample book, and a plurality of image that does not contain target comes the training objective detecting device as negative test sample book.These front samples and negative sample are offered Boosting algorithm actuating unit trains.The Boosting algorithm is well known to a person skilled in the art, does not therefore specifically describe its implementation here.
Through training, choose the strongest Weak Classifier of classification capacity (be designated as T, T be natural number and less than M) constitute object detector as sub-classifier.An object detector example constructed according to the inventor, about 95% positive test sample book and negative test sample book can judge rightly.
As example, can use above method to construct a pedestrian detector.Fig. 3 exemplarily illustrates the position of preceding 20 pairing HOG pieces of sub-classifier of this pedestrian detector.
It will be appreciated by those skilled in the art that the above example that provides only for the purpose of illustration, rather than the present invention will be limited to this.For example, can come the training objective detecting device with the image of different size, can demarcate the front sample and the negative sample of different numbers, object detector can comprise more or less sub-classifier, can choose different features and come training objective detecting device or the like.
The update method of object detector
The update method of object detector is described below in conjunction with specific embodiments and the drawings.According to embodiments of the invention, real-time update object detector in the process of video being carried out online detection.
Application is that video image (hereinafter to be referred as video) to be detected is input in the object detector according to the environment of the method for the renewal object detector of the embodiment of the invention, and manually or automatically determines target from a two field picture.In video was input to process in the object detector continuously, object detector carried out detection and tracking to target.It will be apparent to those skilled in the art that above these processes can realize with various prior aries.
Fig. 4 is the process flow diagram that the method for upgrading object detector according to an embodiment of the invention is shown.The method of this renewal object detector comprises confidence level determining step S401 and step of updating S402.
In confidence level determining step S401, judge whether the confidence level (being also referred to as degree of confidence) of object detector satisfies the scheduled update condition.Usually, when each two field picture is detected, can obtain forming the classification capacity of each sub-classifier of object detector.Can calculate the confidence level of this object detector according to the classification capacity of each sub-classifier.In an embodiment of the present invention, as example, obtain X front sample and Y negative sample from each two field picture, be used to calculate the confidence level of object detector, X and Y are natural numbers.Can use more or less sample to calculate the confidence level of object detector according to the detection requirement of varying level.The method that the confidence level of multiple calculating detecting device is arranged in the prior art, for example, the method of a kind of pedestrian detector's of calculating confidence level is disclosed in the article " Histograms of Oriented Gradients for Human Detection " (http://iselab.cvc.uab.es/files/Publications/2007/PDF/PGV2007.pd f, last visit on August 11st, 2009) of Navneet Dalal and Bill Triggs.It will be appreciated by those skilled in the art that and to use these methods to calculate the confidence level of object detector easily.In addition, can be according to the detection requirement of varying level, artificially or automatically set suitable predetermined threshold.
According to one embodiment of present invention, video is being carried out in the process of online detection, the every detection of object detector finishes a two field picture, just calculates the confidence level of object detector at this two field picture.Judge then whether described confidence level is lower than predetermined threshold.If be lower than predetermined threshold, then upgrade.
According to another embodiment of the present invention, video is being carried out in the process of online detection, judging every the frame of the schedule time or predetermined number whether object detector is lower than predetermined threshold at least one confidence level at preceding frame of present frame and/or next-door neighbour's present frame.As example, check an object detector every 2 frames, judge that object detector is at present frame and/or be close to the former frame of present frame or whether the confidence level of front cross frame is lower than predetermined threshold.As another example,, judge whether described object detector is lower than predetermined threshold at least one confidence level at preceding frame of present frame and/or next-door neighbour's present frame every object detector of 100 milliseconds of detections.If be lower than predetermined threshold, then upgrade.
According to another embodiment of the present invention, described confidence level determining step is configured to video is being carried out in the process of online detection, judges object detector distributes whether meet preassigned pattern at least one confidence level at preceding frame of present frame and/or next-door neighbour's present frame.As example,, think that object detector needs to upgrade when at least one at present frame and/or next-door neighbour's present frame of object detector distributes when presenting downtrending in confidence level of preceding frame.Owing to be that the variation tendency of object detector is judged, rather than judge at single frame, therefore this embodiment make can avoid since the unexpected shake of following the tracks of or a temporary transient tracking fail when causing the instantaneous decline of confidence level of object detector, object detector is carried out unnecessary renewal, and influence the subsequent detection of object detector thus.
Continuation is with reference to figure 4, satisfy the scheduled update condition if in step S401, judge the confidence level of object detector, then in step of updating S402 with the minimum sub-classifier of at least one classification capacity in a plurality of sub-classifiers that comprised in the new sub-classifier replacement object detector.
Fig. 5 illustrates the process flow diagram of step of updating according to an embodiment of the invention.As shown in Figure 5, step of updating S402 may further comprise the steps:
Step S501: at present frame and/or next-door neighbour present frame at least one at preceding frame, from all sub-classifiers of object detector (for example, be T sub-classifier in previously described example) middle N the minimum sub-classifier of classification capacity of removing, N is a natural number, keeps all the other sub-classifiers (for example T-N).
Step S502: utilize the sub-classifier in the alternative sub-classifier pond that at least one of described present frame and/or next-door neighbour's present frame classified at preceding frame, and from alternative sub-classifier pond, choose the highest N of a classification capacity sub-classifier as new sub-classifier.Can choose (for example choosing randomly) W sub-classifier from all available sub-classifiers (for example previously described M sub-classifier), to form described alternative sub-classifier pond, W is a natural number.The step that forms alternative sub-classifier pond can be carried out (for example, carrying out) when step S502 begins in the process of online detection.Also can use from the alternative sub-classifier pond of outside input.
Step S503: utilize described N new all the other T-Ns the sub-classifier of sub-classifier in object detector to form new object detector.
In the above-described embodiments, the sequencing of step S501 and step S502 also can be put upside down.In addition, a number that is appreciated that among above-mentioned steps S501 and the S502 institute's sub-classifier of removing and choosing can determine as required, such as, also can be 1,2,3,4 or more a plurality of.But upgrading the too many words of number may can make the ability of object detector further descend on the contrary, therefore in general upgrades 3 left and right sides sub-classifiers and is advisable in practice.
In one embodiment of the invention, in step S502, before from alternative sub-classifier pond, choosing new sub-classifier, N the sub-classifier that the classification capacity that will remove in step S501 is minimum is put in the alternative sub-classifier pond, so that make this N sub-classifier also participate in choosing subsequently.In addition, in this embodiment,, N the new sub-classifier that selects can be removed from alternative sub-classifier pond in order not make the sub-classifier that duplicates in the alternative sub-classifier pond.
Fig. 6 illustrates the process flow diagram of step of updating according to another embodiment of the present invention.As shown in Figure 6, step of updating S402 may further comprise the steps:
Step S601: at present frame and/or next-door neighbour present frame at least one at preceding frame, from all sub-classifiers of object detector (for example, be T sub-classifier in previously described example) middle 1 the minimum sub-classifier of classification capacity of removing, keep all the other sub-classifiers (for example T-1 sub-classifier).
Step S602: utilize the sub-classifier in the alternative sub-classifier pond that at least one of described present frame and/or next-door neighbour's present frame classified at preceding frame, and from alternative sub-classifier pond, choose 1 sub-classifier that classification capacity is the highest as new sub-classifier.
Step S603: utilize all the other T-1s the sub-classifier of described 1 new sub-classifier in object detector to form new object detector.
Total repeats step S601 to S603 N time, has upgraded the object detector of N sub-classifier with formation.
According to one embodiment of present invention, before chooser sorter from alternative sub-classifier pond, the sub-classifier in the described sub-classifier pond is carried out prescreen.For example, can calculate each sub-classifier in present frame with at least one position in preceding frame of next-door neighbour's present frame on the similarity of color characteristic, and select the higher pairing sub-classifier in position of color similarity degree.In addition, can also check the position at each sub-classifier place, and distribute lower selected weight to the sub-classifier that is positioned at edge and corner.Such process can make the sub-classifier of choosing in the step of updating can put letter more.This prescreen process also can be used in the process that forms alternative sub-classifier pond.
Schematically illustrated 10 the front samples of Fig. 7 A, Fig. 7 B is shown schematically in the diagrammatic sketch of 3 sub-classifiers removing in the step of updating from object detector, and Fig. 7 C is shown schematically in the diagrammatic sketch of 3 new sub-classifiers selecting in the step of updating.In the example shown in Fig. 7 A-7C, be as detecting target with the pedestrian.As seen, when the target pedestrian in the sample bent over, pedestrian's corresponding body part was not in Fig. 7 B on the position of 3 shown sub-classifiers, so these 3 sub-classifiers can't be classified to pedestrian's corresponding body part from Fig. 7 A-7C.In Fig. 7 C, 3 selected new sub-classifiers are in respectively on the position of the current body part of pedestrian, can classify to pedestrian's corresponding body part well.
Fig. 8 illustrates the process flow diagram of the method for renewal object detector according to another embodiment of the present invention.In this embodiment, step S801 and S802 are identical with step of updating S402 with confidence level determining step S401 among the embodiment shown in Figure 4.But, in the embodiment shown in fig. 8, after step of updating S802, judge once more whether the confidence level of object detector satisfies scheduled update condition (step S801), if then repeat described step of updating, with the further object detector that improves.Preferably, after the step of updating of having carried out certain number of times,, then stop this more new technological process, to avoid owing to introducing the robustness that too many new sub-classifier has a strong impact on object detector if the confidence level of object detector still can not meet the demands.
It will be apparent to those skilled in the art that and to carry out various combinations to above-described each embodiment and example within the spirit and scope of the present invention.
The constraint condition of upgrading
To upgrade the risk that causes in order reducing, can under some particular case, to suspend the step of updating in the method for upgrading object detector.
According to one embodiment of present invention, when the color characteristic generation marked change of couple candidate detection target, suspend step of updating.When the color characteristic generation marked change of couple candidate detection target, show, for example, might marked change take place illumination, rather than target itself changes.Therefore, in this case, the reduction of the classification capacity of some sub-classifier has just reflected the variation of described illumination, and does not mean that the classification capacity of these sub-classifiers really reduces.If upgrade object detector this moment, might reduce the detectability of object detector on the contrary.
Here, color characteristic includes but not limited to for example RGB feature, HIS (Hue, Intensity, Saturation) (colourity, brightness, saturation degree) feature, HSV (Hue, Saturation, Value) (colourity, saturation degree, purity) feature etc.For example, can utilize that (Histogram of Color, HC) feature judges whether the change degree of corresponding color characteristic has surpassed predetermined threshold based on for example color histogram in rgb space, HIS space or HSV space.If judge that then marked change has taken place color characteristic.Described predetermined threshold can require and artificially or setting automatically according to actual detected.It will be apparent to those skilled in the art that and to use any color characteristic detection method to detect color characteristic, therefore here it is not specifically described.
According to another embodiment of the invention, when the color characteristic of couple candidate detection target with a low credibility, suspend step of updating.The confidence level of color characteristic represents to utilize this color characteristic can identify the credibility of target.It for example may be because the situation of aforesaid color characteristic generation marked change causes that confidence level reduces, or might follow the tracks of and shake or a temporary transient tracking failure have occurred.Therefore, similarly, upgrade the detectability that sub-classifier might reduce detecting device on the contrary this moment.In addition, those skilled in the art can adopt multiple mode to calculate the confidence level of color characteristic, (it is determined by manual mark for example can to adopt Euclidean distance method, Pasteur (Bhattacharyya) distance method or the like to come the color histogram feature of calculated candidate target and To Template, initial object appearing in the video that perhaps is confirmed as object detector and is detected) therefore the distance between the color histogram feature does not here specifically describe it.
According to another embodiment of the invention, when target and other target is approaching, by other target occlusion or when having blocked other target, suspend step of updating.Those skilled in the art can utilize the whole bag of tricks existing, that developing and that will develop judges above near and the situation of blocking.In addition, target and other target is approaching, by other target occlusion or to have blocked various situations such as other target can be condition from the outside input of embodiments of the present invention.Therefore here not to how judging near specifically describing with blocking.
By carrying out above-mentioned constraint, can make object detector online updating robustness be improved significantly.
Upgrade the device of object detector
The device (hereinafter to be referred as updating device) that upgrades object detector is described below in conjunction with specific embodiments and the drawings.
Fig. 9 is a schematic block diagram of upgrading the device of object detector according to an embodiment of the invention.As shown in Figure 9, updating device 900 comprises renewal judging unit 901 and upgrades performance element 902.Wherein, upgrade judging unit 901 and comprise confidence level judging unit 9011.
According to one embodiment of present invention, confidence level judging unit 9011 is configured to judge whether the confidence level of object detector satisfies the scheduled update condition.Object detector is made up of a plurality of sub-classifiers, is used to detect objects in video.Upgrading performance element 902 is configured to: satisfy the scheduled update condition if confidence level judging unit 9011 is judged confidence level, then replace the minimum sub-classifier of at least one classification capacity in a plurality of sub-classifiers with new sub-classifier.
According to another embodiment of the invention, confidence level judging unit 9011 also is configured to video is being carried out in the process of online detection, during a two field picture in the intact video of the every detection of object detector, judge whether described object detector is lower than predetermined threshold at the confidence level of this two field picture.
According to another embodiment of the invention, confidence level judging unit 9011 also is configured to video be carried out in the process of online detection at object detector, judges every the frame of the schedule time or predetermined number whether object detector is lower than predetermined threshold at least one confidence level at preceding frame of present frame and/or next-door neighbour's present frame.
According to another embodiment of the invention, confidence level judging unit 9011 also is configured to video be carried out in the process of online detection at object detector, judges object detector distributes whether meet preassigned pattern at least one confidence level at preceding frame of present frame and/or next-door neighbour's present frame.
According to another embodiment of the invention, upgrading performance element 902 also is configured to: choose new sub-classifier from alternative sub-classifier pond.As mentioned above, described alternative sub-classifier pond is to be made of the parton sorter of choosing from all available sub-classifiers that obtain in advance.
According to another embodiment of the invention, upgrading performance element 902 also is configured to: choose new sub-classifier from alternative sub-classifier pond before, the minimum sub-classifier of described at least one classification capacity is put back in the alternative sub-classifier pond.
According to another embodiment of the invention, upgrading performance element 902 also is configured to: utilize the sub-classifier in the alternative sub-classifier pond that at least one of present frame or next-door neighbour's present frame classified at preceding frame, and choose the highest sub-classifier of at least one classification capacity as described new sub-classifier from alternative sub-classifier pond.
According to another embodiment of the invention, upgrading performance element 902 also is configured to: before chooser sorter from alternative sub-classifier pond, the sub-classifier in the alternative sub-classifier pond is carried out prescreen.
Figure 10 is the schematic block diagram of the updating device of object detector according to another embodiment of the present invention.Compare with Fig. 9, in updating device shown in Figure 10 1000, upgrade judging unit 1001 and also comprise renewal constraint element 10012.Upgrading constraint element 10012 is configured to described renewal performance element 1002 is not worked: the color characteristic generation marked change of couple candidate detection target; The color characteristic of couple candidate detection target with a low credibility; Perhaps the couple candidate detection target and other target is approaching, by other target occlusion or blocked other target.
The function of the corresponding units among other unit among the embodiment shown in Figure 10 and the embodiment shown in Figure 9 is identical.About the further details of the operation of each unit, can be not described in detail here with reference to each embodiment of the update method of above-described object detector.
Each forms module in the said apparatus, the unit can be configured by the mode of software, firmware, hardware or its combination.Dispose spendable concrete means or mode and be well known to those skilled in the art, do not repeat them here.Under situation about realizing by software or firmware, from storage medium or network the program that constitutes this software is installed to the computing machine with specialized hardware structure (multi-purpose computer 1100 for example shown in Figure 11), this computing machine can be carried out various functions etc. when various program is installed.
In Figure 11, CPU (central processing unit) (CPU) 1101 carries out various processing according to program stored among ROM (read-only memory) (ROM) 1102 or from the program that storage area 1108 is loaded into random-access memory (ram) 1103.In RAM 1103, also store data required when CPU 1101 carries out various processing or the like as required.CPU 1101, ROM 1102 and RAM 1103 are connected to each other via bus 1104.Input/output interface 1105 also is connected to bus 1104.
Following parts are connected to input/output interface 1105: importation 1106 (comprising keyboard, mouse or the like), output 1107 (comprise display, such as cathode ray tube (CRT), LCD (LCD) etc. and loudspeaker etc.), storage area 1108 (comprising hard disk etc.), communications portion 1109 (comprising that network interface unit is such as LAN card, modulator-demodular unit etc.).Communications portion 1109 is handled such as the Internet executive communication via network.As required, driver 1110 also can be connected to input/output interface 1105.Detachable media 1111 can be installed on the driver 1110 as required such as disk, CD, magneto-optic disk, semiconductor memory or the like, makes the computer program of therefrom reading be installed to as required in the storage area 1108.
Realizing by software under the situation of above-mentioned series of processes, such as detachable media 1111 program that constitutes software is being installed such as the Internet or storage medium from network.
It will be understood by those of skill in the art that this storage medium is not limited to shown in Figure 11 wherein having program stored therein, distribute separately so that the detachable media 1111 of program to be provided to the user with equipment.The example of detachable media 1111 comprises disk (comprising floppy disk (registered trademark)), CD (comprising compact disc read-only memory (CD-ROM) and digital universal disc (DVD)), magneto-optic disk (comprising mini-disk (MD) (registered trademark)) and semiconductor memory.Perhaps, storage medium can be hard disk that comprises in ROM 1102, the storage area 1108 or the like, computer program stored wherein, and be distributed to the user with the equipment that comprises them.
The present invention also proposes a kind of program product that stores the instruction code that machine readable gets.When described instruction code is read and carried out by machine, can carry out above-mentioned method according to the embodiment of the invention.
Correspondingly, being used for carrying the above-mentioned storage medium that stores the program product of the instruction code that machine readable gets is also included within of the present invention open.Described storage medium includes but not limited to floppy disk, CD, magneto-optic disk, storage card, memory stick or the like.
In the above in the description to the specific embodiment of the invention, can in one or more other embodiment, use in identical or similar mode at the feature that a kind of embodiment is described and/or illustrated, combined with the feature in other embodiment, or the feature in alternative other embodiment.
Should emphasize that term " comprises/comprise " existence that refers to feature, key element, step or assembly when this paper uses, but not get rid of the existence of one or more further feature, key element, step or assembly or additional.
In addition, the time sequencing of describing during method of the present invention is not limited to is to specifications carried out, also can according to other time sequencing ground, carry out concurrently or independently.Therefore, the execution sequence of the method for describing in this instructions is not construed as limiting technical scope of the present invention.
Although the present invention is disclosed above by description to specific embodiments of the invention,, should be appreciated that all above-mentioned embodiment and example all are illustrative, and not restrictive.Those skilled in the art can design various modifications of the present invention, improvement or equivalent in the spirit and scope of claims.These modifications, improvement or equivalent also should be believed to comprise in protection scope of the present invention.
Claims (19)
1. method of upgrading object detector, described object detector is made up of a plurality of sub-classifiers, is used to detect objects in video, and described method comprises:
The confidence level determining step judges whether the confidence level of described object detector satisfies the scheduled update condition; And
Step of updating if described confidence level satisfies the scheduled update condition, then replaces the minimum sub-classifier of at least one classification capacity in described a plurality of sub-classifier with new sub-classifier.
2. the method for claim 1, wherein, described confidence level determining step comprises: during a two field picture in the intact video of the every detection of object detector, whether the confidence level of judging described object detector satisfies the scheduled update condition, and what described scheduled update condition was described object detector at this two field picture is with a low credibility in predetermined threshold.
3. the method for claim 1, wherein, described confidence level determining step comprises: whether the confidence level of judging described object detector every the frame of the schedule time or predetermined number satisfies the scheduled update condition, and described scheduled update condition is described object detector at a low credibility in predetermined threshold at preceding frame of at least one of present frame and/or next-door neighbour's present frame.
4. to be described object detector at least one confidence level at preceding frame of present frame and/or next-door neighbour's present frame distribute the method for claim 1, wherein described scheduled update condition meets preassigned pattern.
5. the method for claim 1, wherein described new sub-classifier is chosen from the sub-classifier pond, and described sub-classifier pond is to be made of the parton sorter of choosing from all available sub-classifiers that obtain in advance.
6. method as claimed in claim 5 also comprises:
Before from the sub-classifier pond, choosing described new sub-classifier, the minimum sub-classifier of described at least one classification capacity is put back in the sub-classifier pond.
7. method as claimed in claim 5 also comprises:
Utilize the sub-classifier in the described sub-classifier pond that at least one of present frame or next-door neighbour's present frame classified at preceding frame, and from described sub-classifier pond, choose the highest sub-classifier of at least one classification capacity as described new sub-classifier.
8. the method for claim 1 also comprises:
After described step of updating, judge once more whether the confidence level of described object detector satisfies the scheduled update condition, if then repeat described step of updating.
9. the method for claim 1, wherein in following situation, do not carry out step of updating in any case: the color characteristic generation marked change of couple candidate detection target; The color characteristic of couple candidate detection target with a low credibility; Perhaps couple candidate detection target and other object is approaching, blocked or blocked other object by other object.
10. method as claimed in claim 5 also comprises:
Before chooser sorter from described sub-classifier pond, the sub-classifier in the described sub-classifier pond is carried out prescreen.
11. a device that upgrades object detector, described object detector is made up of a plurality of sub-classifiers, is used to detect objects in video, and described device comprises:
Upgrade judging unit, described renewal judging unit comprises the confidence level judging unit, and described confidence level judging unit is configured to judge whether the confidence level of described object detector satisfies the scheduled update condition; And
Upgrade performance element, be configured to:, then replace the minimum sub-classifier of at least one classification capacity in described a plurality of sub-classifier with new sub-classifier if described confidence level satisfies the scheduled update condition.
12. device as claimed in claim 11, wherein, when described confidence level judging unit also is configured to a two field picture in the intact video of the every detection of object detector, whether the confidence level of judging described object detector satisfies the scheduled update condition, and what described scheduled update condition was described object detector at this two field picture is with a low credibility in predetermined threshold.
13. device as claimed in claim 11, wherein, described confidence level judging unit also is configured to judge every the frame of the schedule time or predetermined number whether the confidence level of described object detector satisfies the scheduled update condition, and described scheduled update condition is described object detector at a low credibility in predetermined threshold at preceding frame of at least one of present frame and/or next-door neighbour's present frame.
14. device as claimed in claim 11, wherein, to be described object detector at least one confidence level at preceding frame of present frame and/or next-door neighbour's present frame distribute described scheduled update condition meets preassigned pattern.
15. device as claimed in claim 11, wherein, described renewal performance element also is configured to: choose described new sub-classifier from the sub-classifier pond, described sub-classifier pond is to be made of the parton sorter of choosing from all available sub-classifiers that obtain in advance.
16. device as claimed in claim 15, wherein said renewal performance element also is configured to: choose described new sub-classifier from the sub-classifier pond before, the minimum sub-classifier of described at least one classification capacity is put back in the sub-classifier pond.
17. device as claimed in claim 15, wherein, described renewal performance element also is configured to: utilize the sub-classifier in the described sub-classifier pond that at least one of present frame or next-door neighbour's present frame classified at preceding frame, and choose the highest sub-classifier of at least one classification capacity as described new sub-classifier from described sub-classifier pond.
18. device as claimed in claim 11, wherein, described renewal judging unit also comprises:
Upgrade constraint element, be configured to described renewal performance element is not worked: the color characteristic generation marked change of couple candidate detection target; The color characteristic of couple candidate detection target with a low credibility; Perhaps couple candidate detection target and other object is approaching, blocked or blocked other object by other object.
19. device as claimed in claim 15, wherein, described renewal performance element also is configured to: before chooser sorter from described sub-classifier pond, the sub-classifier in the described sub-classifier pond is carried out prescreen.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910166494.XA CN101996400B (en) | 2009-08-19 | 2009-08-19 | Upgrade the method and apparatus of object detector |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200910166494.XA CN101996400B (en) | 2009-08-19 | 2009-08-19 | Upgrade the method and apparatus of object detector |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101996400A true CN101996400A (en) | 2011-03-30 |
| CN101996400B CN101996400B (en) | 2015-09-09 |
Family
ID=43786521
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN200910166494.XA Expired - Fee Related CN101996400B (en) | 2009-08-19 | 2009-08-19 | Upgrade the method and apparatus of object detector |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101996400B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831442A (en) * | 2011-06-13 | 2012-12-19 | 索尼公司 | Abnormal behavior detection method and equipment and method and equipment for generating abnormal behavior detection equipment |
| CN103425991A (en) * | 2012-05-15 | 2013-12-04 | 富士通株式会社 | Method and device for classifying targets in video |
| CN104599287A (en) * | 2013-11-01 | 2015-05-06 | 株式会社理光 | Object tracking method and device and object recognition method and device |
-
2009
- 2009-08-19 CN CN200910166494.XA patent/CN101996400B/en not_active Expired - Fee Related
Non-Patent Citations (6)
| Title |
|---|
| HELMUT GRABNER AND HORST BISCHOF: "On-line Boosting and Vision", 《2006 IEEE COMPUTER SOCIETY CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| HELMUT GRABNERET ET AL: "Real-Time Tracking via On-line Boosting", 《PROCEEDINGS OF THE BRITISH MACHINE VISION CONFERENCE》 * |
| NAVNEET DALAL AND BILL TRIGGS: "Histograms of Oriented Gradients for Human Detection", 《PROCEEDINGS OF THE 2005 IEEE COMPUTER SOCIETY CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| SHAI AVIDAN: "Ensemble Tracking", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 崔潇潇 等: "基于级联Adaboost的目标检测融合算法", 《自动化学报》 * |
| 王路 等: "基于Co-Training的协同目标跟踪", 《计算机工程》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102831442A (en) * | 2011-06-13 | 2012-12-19 | 索尼公司 | Abnormal behavior detection method and equipment and method and equipment for generating abnormal behavior detection equipment |
| CN103425991A (en) * | 2012-05-15 | 2013-12-04 | 富士通株式会社 | Method and device for classifying targets in video |
| CN104599287A (en) * | 2013-11-01 | 2015-05-06 | 株式会社理光 | Object tracking method and device and object recognition method and device |
| CN104599287B (en) * | 2013-11-01 | 2018-01-16 | 株式会社理光 | Method for tracing object and device, object identifying method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101996400B (en) | 2015-09-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP1745414B1 (en) | Method for combining boosted classifiers for efficient multi-class object detection | |
| CN102693432B (en) | Use reliable partial model more to newly arrive and regulate clear path to detect | |
| US20170032247A1 (en) | Media classification | |
| CN104134078B (en) | Automatic selection method for classifiers in people flow counting system | |
| CN101877064B (en) | Image classification method and image classification device | |
| EP2763078A1 (en) | Method, apparatus and computer readable recording medium for detecting a location of a face feature point using an adaboost learning algorithm | |
| Wang et al. | Automated classification of metaphase chromosomes: optimization of an adaptive computerized scheme | |
| US20220277174A1 (en) | Evaluation method, non-transitory computer-readable storage medium, and information processing device | |
| CN110222582B (en) | Image processing method and camera | |
| US20190156125A1 (en) | Characterizing Content with a Predictive Error Representation | |
| US11600088B2 (en) | Utilizing machine learning and image filtering techniques to detect and analyze handwritten text | |
| CN104182722A (en) | Text detection method and device and text information extraction method and system | |
| Jung et al. | Support vector number reduction: Survey and experimental evaluations | |
| JP6947005B2 (en) | Attribute recognition device, attribute recognition method, and machine learning device | |
| García et al. | Supervised texture classification by integration of multiple texture methods and evaluation windows | |
| CN101996400B (en) | Upgrade the method and apparatus of object detector | |
| Wali et al. | A new system for event detection from video surveillance sequences | |
| CN116348924A (en) | Anomaly detector for detecting anomalies using a complementary classifier | |
| CN113496251A (en) | Device for determining a classifier for identifying an object in an image, device for identifying an object in an image and corresponding method | |
| JP5389723B2 (en) | Object detection device and learning device thereof | |
| US9619521B1 (en) | Classification using concept ranking according to negative exemplars | |
| Holmes | A genetics-based machine learning approach to knowledge discovery in clinical data | |
| KR102050422B1 (en) | Apparatus and method for recognizing character | |
| Torralba et al. | Shared features for multiclass object detection | |
| EP2093709A1 (en) | Document image feature value generating device, document image feature value generating method, and document image feature value generating program |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20150909 |