CN105095848B - The method and system of target identification - Google Patents
The method and system of target identification Download PDFInfo
- Publication number
- CN105095848B CN105095848B CN201410720074.2A CN201410720074A CN105095848B CN 105095848 B CN105095848 B CN 105095848B CN 201410720074 A CN201410720074 A CN 201410720074A CN 105095848 B CN105095848 B CN 105095848B
- Authority
- CN
- China
- Prior art keywords
- classifier
- default
- detection
- testing result
- layer
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method and system of target identification.This method comprises the following steps: using in default detection of classifier given image whether the default local feature containing object, and obtain testing result；According to testing result, if so, determining to include object in given image；If it is not, then determining not including object in given image.Wherein, default local feature is the feature for the component home different from other objects in addition to object for including in selected object, and component home is a part of object.Its preset it needs to be determined that target most separating capacity component home, target identification has a specific aim very much, and recognition efficiency is high, and it is more accurate to identify.And avoid target identification operand in traditional technology big using component home, the inadequate problem of accuracy rate.
Description
Technical field
The present invention relates to image identification technical field more particularly to a kind of method and system of target identification.
Background technique
Object identification is primarily referred to as the fixation and recognition on given image and goes out preassigned object.Know in object
Other field is mainly concerned with following three kinds of algorithms:
1. template matching method
Template matching method main thought is that object to be matched is whole or a part is as template, in image to be searched
Upper sequence takes out image block identical with template size by certain strategy, and it is similar to above-mentioned image block then to compare template
Degree, similarity-rough set use Euclidean distance or other algorithm.If template is less than some threshold value at a distance from image block, recognize
It is exactly the object found for the image block, target identification is completed.This method is simple, it is easy to accomplish.But have simultaneously very much
Disadvantage:
1) operand is big, and needing to do many optimizations just may be implemented in real time.
2) low precision can not correspond to the variation of scale and angle.Object in image slightly change size or
Visual angle just cannot identify.
3) it is big to occupy storage space, for the object being recognized accurately under various illumination conditions and view transformation, needs
The template of the object in various situations is stored, a large amount of storage space can be occupied.
2. Feature Points Matching method
The main thought of Feature Points Matching method is to be detected on target object image using certain algorithm (such as SIFT algorithm)
Then multiple characteristic points that can represent the object out extract feature vector (such as gradient histogram around these characteristic points
Figure) for identification later.Multiple characteristic points are detected using same procedure in images to be recognized, then extract characteristic point
The feature vector of surrounding.It is compared with the feature vector of these feature vectors and object, if distance is less than certain threshold value
Then think to find target, identification process is completed.
Target can be also recognized accurately in the case where the angle of object and dimensional variation in the algorithm.But the algorithm is only
It is abundant to be used for profile, the object of rigid plane deformation, such as book cover, trade mark etc., if because illumination variation causes to take turns
Exterior feature, which is not known, will affect recognition correct rate.Another problem is the computationally intensive of characteristic point detection.
3. the method based on statistical learning
It is all used widely based on the method for statistical learning in every field.The main thought of this method is to collect largely
The image of object to be identified is collected simultaneously the image of a large amount of non-objects as negative sample as positive sample.To positive and negative sample
Then this extraction feature is trained using data of the method for machine learning to positive negative sample, in feature space study to one
Kind division methods can distinguish positive negative sample, and this method is referred to as classifier, can distinguish positive negative sample.Rank is trained at this time
Section finishes.In service stage, feature is extracted to images to be recognized in the same way, is then input to point that the training stage obtains
In class device, judge whether the image is object according to calculated score.
Greater advantage is embodied in robustness everyway based on the method for statistical learning, but in specific implementation process
In, and have various skills and method.At present using it is wide be Bag Of Words model, these methods have some shortcomings
Place:
3.1. higher for two classification problem accuracy, but for more classification problems with the increase of classification, accuracy
Sharply decline.
3.2. due to only extracting global characteristics, for the macroscopically two kinds biggish objects of difference, discrimination is very high.But it is right
In generally much like, only the discrimination of two kinds of different objects of local detail is very low.
3.3. for object is in the image under complex background, study is easy in training learning process and is arrived and mesh
It marks in unrelated background, and reduces the feature of object, reduce discrimination.
In conclusion seeking a kind of high efficiency and target identification method with high accuracy is a kind of urgent problem to be solved.
Summary of the invention
Based on this, it is necessary to which computationally intensive for target identification in traditional technology, the not high problem of accuracy provides one kind
The method and system for the target identification that target identification is high-efficient and recognition accuracy is high.
A kind of method for target identification that purpose provides to realize the present invention, comprising the following steps:
Using in default detection of classifier given image whether the default local feature containing object, and obtain detection knot
Fruit；
According to the testing result, if so, determining in the given image comprising the object；
According to the testing result, if it is not, then determining not including the object in the given image；
Wherein, the default local feature is to include in the selected object different from addition to the object
Other objects component home feature, and the component home be the object a part.
The embodiment of a kind of method as target identification, the default classifier include N layers, respectively first layer
Default classifier, the second layer preset classifier ... ..., and n-th layer presets classifier, and the quantity of every layer of classifier is at least two
It is a；
It is described using in default detection of classifier given image whether the default local feature containing object, and examined
Survey result, comprising the following steps:
Firstly, presetting classifier using first layer carries out first layer classification to the given image；
Classify secondly, continuing the second layer using corresponding second layer classifier according to the result of first layer classification；
……；
Until using default local feature whether is contained in given image described in corresponding n-th layer detection of classifier, and obtain
To testing result；
Wherein, N is natural number, and is more than or equal to 2.
The embodiment of a kind of method as target identification, the quantity of the default classifier and the target to be detected
The quantity of object is identical, and a kind of each default object of detection of classifier；
It is described using in default detection of classifier given image whether the default local feature containing object, and examined
Survey result, comprising the following steps:
Default local feature is carried out to the given image using each default classifier to detect, and provide it is described to
Determine in image whether include the default local feature scoring；
Judge whether the scoring is greater than default score value, if so, obtaining the testing result for "Yes"；
If it is not, then obtaining the testing result for "No".
The embodiment of a kind of method as target identification, when there are two the default classifier of the above obtain for
When the testing result of "Yes", determines that the default classifier of highest scoring obtains the testing result for "Yes", determine that remaining is obtained
Dividing is not that the highest default classifier obtains the testing result for "No".
The embodiment of a kind of method as target identification, the local feature are to carry out spy to the component home
When levying what extraction obtained, and carrying out feature extraction to the component home, the component home is divided into W × H block, and right
Each block takes the gradient orientation histogram of 9 buckets (bin)；
Wherein W and H is positive integer.
The embodiment of a kind of method as target identification, the default classifier in the training process, positive sample
The local feature of each form including the object, negative sample are the local background's feature and other correlations of the object
The feature of object.
A kind of system of target identification based on the same inventive concept, including detection module, the first determination module and second
Determination module, in which:
The detection module, for use in default detection of classifier given image whether the default part containing object
Feature, and obtain testing result；
First determination module, for the testing result according to the detection module, if so, determining the given figure
It include the object as in；
Second determination module, for the testing result according to the detection module, if it is not, then determining the given figure
The object is not included as in；
Wherein, the default local feature is to include in the selected object different from addition to the object
Other objects component home feature, and the component home be the object a part.
A kind of embodiment of system as target identification, the default classifier include N layers, respectively first layer
Default classifier, the second layer preset classifier ... ..., and n-th layer presets classifier, and the quantity of every layer of classifier is at least two
It is a；
The detection module includes N number of detection monitoring submodule, respectively the first detection sub-module, the second detection submodule
Block ... ..., N detection sub-module, the detection module when being detected,
Firstly, the first detection sub-module, which presets classifier using first layer, carries out first layer classification to the given image；
Secondly, the second detection sub-module uses the corresponding second layer according to the first layer classification results of the first detection sub-module
Classifier continues second layer classification；
……；
Until whether N detection sub-module is default using containing in given image described in corresponding n-th layer detection of classifier
Local feature, and obtain testing result；
Wherein, N is natural number, and is more than or equal to 2.
A kind of embodiment of system as target identification, the quantity of the default classifier and the target to be detected
The quantity of object is identical, and a kind of each default object of detection of classifier；
The detection module carries out default local feature to the given image using each default classifier and detects
When, provide in the given image whether include the default local feature scoring；
It further include judging submodule in the detection module, for judging whether the scoring is greater than default score value, if
It is then to obtain the testing result for "Yes"；
If it is not, then obtaining the testing result for "No".
A kind of embodiment of system as target identification further includes screening submodule in the detection module, uses
In when obtaining the testing result for "Yes" there are two the default classifier of the above, the default classification of highest scoring is determined
Device obtains the testing result for "Yes", and determining remaining score not is that the highest default classifier obtains the detection knot for "No"
Fruit.
The beneficial effect comprise that a kind of method and system of target identification provided by the invention, by setting
It is fixed it needs to be determined that target most separating capacity component home, such as two closely similar but only different automobiles of car light,
Select component home of the car light as target identification.This target identification method has specific aim very much, makes target identification efficiency
Height, and identification is more accurate.And with clearly defined objective target identification and differentiation are carried out using component home and avoids mesh in traditional technology
It is big to identify other operand, the inadequate problem of accuracy rate.
Detailed description of the invention
Fig. 1 is a kind of flow chart of a specific embodiment of the method for target identification of the present invention；
Fig. 2 is a kind of schematic diagram of the layering identification object of a specific embodiment of the method for target identification of the present invention；
Fig. 3 is a kind of identification process schematic diagram of a specific embodiment of the method for target identification of the present invention；
Fig. 4 is that a kind of component home block of a specific embodiment of the method for target identification of the present invention divides schematic diagram；
Fig. 5 is that a kind of positive and negative samples of a specific embodiment of the method for target identification of the present invention acquire schematic diagram；
Fig. 6 to Fig. 8 is to preset classifier training process schematic in a kind of method of target identification of the present invention；
Fig. 9 is a kind of system construction drawing of a specific embodiment of the system of target identification of the present invention；
Figure 10 is that a kind of detection module of a specific embodiment of the system of target identification of the present invention constitutes schematic diagram；
Figure 11 is that a kind of detection module of the another specific embodiment of the system of target identification of the present invention constitutes schematic diagram.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to of the invention
The specific embodiment of the method and system of target identification is illustrated.It should be appreciated that specific embodiment described herein is only
Only to explain the present invention, it is not intended to limit the present invention.
The method of the target identification of one embodiment of the invention, as shown in Figure 1, comprising the following steps:
S100, using in default detection of classifier given image whether the default local feature containing object, and obtain
Testing result.
S200, according to the testing result, if so, determining in the given image comprising the object.
S300, according to the testing result, if it is not, then determining not including the object in the given image.
Wherein, the default local feature is to include in the selected object different from addition to the object
Other objects component home feature, and the component home be the object a part.
Herein it should be noted that in the embodiment of the present invention, component home is target that is preset, manually screening out
Most representative part under each form of object.It is obtained in the default classifier comprising carrying out feature extraction to component home
Local feature.And default classifier includes multiple classifiers under normal circumstances, especially similar in the multiple entirety of differentiation, part
It include specific distinguishing feature in different classifications device, so as to quickly will be similar when differentiated automobile and other items
Product zone separate.Ratio is integrally identified with product, greatly improves the efficiency of target identification.
The method of the target identification of the embodiment of the present invention, by preset it needs to be determined that target most separating capacity
Component home, such as two closely similar but only different automobiles of car light, part of the artificial selection car light as target identification
Component.This target identification method takes full advantage of the priori knowledge of people, and has very much specific aim, makes target identification efficiency
Height, it is more accurate to identify.And the identification and differentiation of target are carried out using component home, avoid target identification operation in traditional technology
Amount is big, the inadequate problem of accuracy rate.
Preferably, when simultaneously identify multiple target products when, settable multiple default classifiers, to the target to be identified into
The identification step by step of row hierarchical classification.Setting default classifier includes N layers, and respectively it is default to preset classifier, the second layer for first layer
Classifier ... ..., n-th layer presets classifier, and the quantity of every layer of classifier is at least two, that is, every layer will at least give
Multiple images be divided into two classes.Certainly multiclass can also be divided on a certain layer as needed.
Wherein, the division of level is classified to the component home of the object screened, and similar component is divided into
One kind, to obtain the biggish major class of several discriminations.Then is obtained being individually trained classification to the group in major class
Two layers of classifier, and the rest may be inferred, is sub-divided into third layer classifier, the 4th layer of classifier ... ..., n-th layer according to the actual situation
Classifier.
Step S100, using in default detection of classifier given image whether the default local feature containing object, and
Obtain testing result, comprising the following steps:
Firstly, presetting classifier using first layer carries out first layer classification to given image.Secondly, being classified according to first layer
Result using corresponding second layer classifier continue the second layer classify.And successively classify ... ... (successively classification)；
Until using default local feature whether is contained in given image described in corresponding n-th layer detection of classifier, and obtain detection knot
Fruit.
Wherein, N is natural number, and is more than or equal to 2.
It is further detailed below with a specific example.
Referring to fig. 2, classified first using first layer classifier to six images given in first layer, six are schemed
As being divided into two groups (as shown in second layers in figure).Every group of image in the second layer is used respectively in second layer classifier
Corresponding default classifier identified that as shown in Figure 2, right part Direct Recognition has gone out object, and left side is into one
Step has obtained two groups of images of third layer.Classifier is preset using corresponding respectively similarly for two groups of images of left side third layer
Carry out further object identification, obtain the 4th layer shown in final recognition result.
Object is successively identified using Multilayer Classifier in the embodiment of the present invention.First layer is preset classifier and is used for
Identify the biggish objects of multiple differences, second layer classifier multiple objects for having certain similarity for identification, third layer
And so on, until finally identifying object.The identification process of multi-class targets object is accelerated by the way of Classification and Identification.
In the wherein embodiment of the method for a target identification, the quantity of the default classifier and the target to be detected
The quantity of object is identical, and a kind of each default object of detection of classifier.At this point, step S100, uses default point
Class device detect in given image whether the default local feature containing object, and obtain testing result, comprising the following steps:
S101 carries out default local feature to the given image using each default classifier and detects, and provides
In the given image whether include the default local feature scoring.
S102, judges whether the scoring is greater than default score value, if so, obtaining the testing result for "Yes"；If it is not,
Then obtain the testing result for "No".
Score value is preset herein, can be carried out according to accuracy of default classifier during carrying out object recognition training
Setting.
In the embodiment of the present invention, while when identifying multiple objects, the corresponding classifier of each object, for one
Given image carries out object identification using each default classifier, obtains recognition result.
Further, when obtaining the testing result for "Yes" there are two the default classifier of the above, determine score most
The high default classifier obtains the testing result for "Yes", and determining remaining score not is that the highest default classifier obtains
To the testing result for "No".
Herein it should be noted that carrying out testing result optimization (score identification) herein is to be directed to the same given image,
And the given image is given images to be recognized, the images to be recognized be obtained from the data such as video sequence with it is preparatory
The image that the image size of component home under each form of the object of storage matches.Images to be recognized can be used software from
It is obtained automatically in the data such as video sequence.
As shown in figure 3, for the process identified to an images to be recognized.It include three default classification in this example
Device, respectively can identification component 1 classifier (recognition result be object 1), can identification component 2 classifier (recognition result
For object 2) and can identification component 3 classifier (recognition result is object 3).Three default classifiers are respectively to be identified
Image is identified, and respectively obtains score 1, score 2 and score 3.Assuming that three scores all reach default score value, then into one
Step is compared three scores, obtains score highest one as final recognition result.If score 1 is 60, score 2 is
70, and score 3 is 95, it is concluded that images to be recognized is the component home image of object 3.
It should be noted that can directly determine wait know when score obviously too low (lower than default score value) there are two if
Other image is the recognition result of a high classifier of score.
Final recognition result is limited by the way of scoring in the embodiment of the present invention, further increases target identification
Precision.Keep the precision of local feature recognition higher, so that it is small to difference more so that method of the invention is preferably used for
The identification of a object.
Herein it should be noted that when carrying out layering identification, the quantity of default classifier should be greater than object to be identified
Quantity, this field be readily appreciated that its also need include carry out Classification and Identification default classifier.But no matter which kind of uses
Mode requires to preset classifier correspondingly with each object and finally identifies object.
Wherein, the local feature obtains component home progress feature extraction, and to the component home
When carrying out feature extraction, as shown in figure 4, the component home is divided into W (width) × H (height) a block, and 9 are taken to each block
The gradient orientation histogram of a bucket (bin) shares 9 × W × H dimension feature hence for a component home.Wherein W and H are equal
For positive integer.
And the default classifier is in the training process, positive sample includes that the part of each form of the object is special
Sign, negative sample are local background's feature of the object and the feature of other related items.And the training of data uses
Adaboost classifier or SVM classifier.
Illustrate the training process of default classifier with the example of a specific vehicle cab recognition below.
In order to identify three kinds of vehicles, it is car light that the stabilizing local component of three kinds of vehicles can be distinguished by, which selecting, and car light is marked
For positive sample, stochastical sampling gone out car light using for region as negative sample.Vehicle is just had identified if having identified car light.
The acquisition of positive negative sample is as shown in Figure 5.First behavior object to be identified: vehicle 1, vehicle 2 and vehicle 3.It is left in second row
Three icons of side are positive sample, and respectively component home-component 1 of vehicle 1, the corresponding component 2 of vehicle 2 and vehicle 3 is corresponding
Component 3.And right side is three negative samples of stochastical sampling respectively.In order to identifying three above-mentioned components, need for this 3
3 classifiers of a component training.Trained process difference is as shown in Figure 6 to 8.In the classifier of training energy identification component 1,
Part component 1 is used as positive sample, carries out classifier instruction for all negative samples shown in component 2, component 3 and Fig. 5 as negative sample
Practice.The classifier of energy identification component 1 at final training.Identify that component 1 also just has identified vehicle 1, and so training can be quasi-
True exclusion vehicle 2 and vehicle 3, recognition accuracy are high.Training for the classifier of energy identification component 2 and energy identification component 3
It is similar with the energy training of classifier of identification component 1, it can refer to attached drawing 7 and attached drawing 8 carry out, details are not described herein again.
It should be noted that positive sample only lists a kind of office of the component under form when this is in citing and is illustrated
The local feature image of object and component under variform may be selected as just in portion's characteristics of image during hands-on
Sample so can reach the effect that object can be identified under variform.
Based on the same inventive concept, the embodiment of the present invention provides a kind of system of target identification, since the solution of this system is asked
The principle of topic is similar to a kind of aforementioned method of target identification, and therefore, the implementation of the system can be according to the specific of preceding method
Step realizes that overlaps will not be repeated.
The system of the target identification of one embodiment of the invention, as shown in figure 9, including detection module 100, the first determination module
200 and second determination module 300.Wherein: whether detection module 100 contains for using in default detection of classifier given image
There is the default local feature of object, and obtains testing result；First determination module 200, for according to the detection module
Testing result, if so, determining in the given image comprising the object；Second determination module 300, for according to
The testing result of detection module, if it is not, then determining not including the object in the given image.
Wherein, the default local feature is to include in the selected object different from addition to the object
Other objects component home feature, and the component home be the object a part.
In the embodiment of the present invention, component home be under each form of object that is pre-selected, manually screening out most
Representative part.Comprising carrying out the local feature that feature extraction obtains to component home in the default classifier.And it is pre-
It is especially similar in the multiple entirety of differentiation if classifier includes multiple classifiers under normal circumstances, the local apparent automobile of difference
And other items when, include specific distinguishing feature in different classifications device, so as to quickly separate similar product zone.
Ratio is integrally identified with product, greatly improves the efficiency of target identification.
The system of the target identification of the embodiment of the present invention, by preset it needs to be determined that target most separating capacity
Component home, such as two closely similar but only different automobiles of car light select component home of the car light as target identification.
This target identification method takes full advantage of the priori knowledge of people, and has very much specific aim, keeps target identification high-efficient, identifies
It is more accurate.And the identification and differentiation of target are carried out using component home, and avoid target identification operand in traditional technology big, it is quasi-
The inadequate problem of true rate.
In the wherein implementation of the system of a target identification, the default classifier includes N layers, and respectively first layer is pre-
If classifier, the second layer preset classifier ... ..., n-th layer presets classifier, and the quantity of every layer of classifier is at least two.
The detection module 100 includes that N number of detection monitors submodule, as shown in Figure 10, respectively the first detection sub-module
101, the second detection sub-module 102 ... ..., N detection sub-module 10N, the detection module when being detected,
Firstly, the first detection sub-module 101, which presets classifier using first layer, carries out first layer point to the given image
Class；
Secondly, the second detection sub-module 102 uses corresponding the according to the first layer classification results of the first detection sub-module
Two layers of classifier continue second layer classification；
……；
Until whether N detection sub-module 10N is pre- using containing in given image described in corresponding n-th layer detection of classifier
If local feature, and obtain testing result.
Wherein, N is natural number, and is more than or equal to 2.
Object is successively identified using Multilayer Classifier in the embodiment of the present invention.First layer is preset classifier and is used for
Identify the biggish objects of multiple differences, second layer classifier multiple objects for having certain similarity for identification, third layer
And so on, until finally identifying object.The identification process of multi-class targets object is accelerated by the way of Classification and Identification.
In the embodiment of the system of another target identification, as shown in figure 11, the quantity of the default classifier with to examine
The quantity of the object (object to be detected) of survey is identical, and a kind of each default object of detection of classifier.
Detection module 100 carries out default local feature to the given image using each default classifier and detects
When, provide in the given image whether include the default local feature scoring
And further include judging submodule 110 in detection module 100, for judging whether the scoring is greater than default scoring
Value, if so, obtaining the testing result for "Yes"；If it is not, then obtaining the testing result for "No".
It further include screening submodule 120 in further detection module 100, for when there are two default classification described above
When device obtains the testing result for "Yes", determine that the default classifier of highest scoring obtains the testing result for "Yes", really
Remaining fixed score is not that the highest default classifier obtains the testing result for "No".So that it is determined that given image is score
The corresponding object of highest classifier.
Final recognition result is limited by the way of scoring in the embodiment of the present invention, further increases target identification
Precision.Keep the precision of local feature recognition higher, so that it is small to difference more so that method of the invention is preferably used for
The identification of a object.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
Limitations on the scope of the patent of the present invention therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to guarantor of the invention
Protect range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (8)
1. a kind of method of target identification, which comprises the following steps:
Using in default detection of classifier given image whether the default local feature containing object, and obtain testing result；
According to the testing result, if so, determining in the given image comprising the object；
According to the testing result, if it is not, then determining not including the object in the given image；
Wherein, the default local feature is its in addition to the object of being different from for including in the selected object
The feature of the component home of his object, and the component home is a part of the object；
The quantity of the default classifier is identical as the quantity for the object to be detected, and each default detection of classifier one
The kind object；
It is described using in default detection of classifier given image whether the default local feature containing object, and obtain detection knot
Fruit, comprising the following steps:
Default local feature is carried out to the given image using each default classifier to detect, and provides the given figure
As in whether include the default local feature scoring；
Judge whether the scoring is greater than default score value, if so, obtaining the testing result for "Yes"；
If it is not, then obtaining the testing result for "No".
2. the method for target identification according to claim 1, which is characterized in that the default classifier includes N layers, respectively
Classifier to n-th layer is preset for first layer and presets classifier, and the quantity of every layer of classifier is at least two；
It is described using in default detection of classifier given image whether the default local feature containing object, and obtain detection knot
Fruit, comprising the following steps:
Firstly, presetting classifier using first layer carries out first layer classification to the given image；
Classify secondly, continuing the second layer using corresponding second layer classifier according to the result of first layer classification；
And so on；
Until using the default local feature whether is contained in given image described in corresponding n-th layer detection of classifier, and obtain
To testing result；
Wherein, N is natural number, and is more than or equal to 2.
3. the method for target identification according to claim 1, which is characterized in that when there are two the default classifiers of the above
It when obtaining the testing result for "Yes", determines that the default classifier of highest scoring obtains the testing result for "Yes", determines
Remaining score is not that the highest default classifier obtains the testing result for "No".
4. the method for target identification according to claim 1, which is characterized in that the local feature is to the part portion
Part carries out feature extraction and obtains, and when carrying out feature extraction to the component home, the component home is divided into W × H
Block, and the gradient orientation histogram of 9 buckets is taken to each block；
Wherein W and H is positive integer.
5. the method for target identification according to claim 1, which is characterized in that the default classifier is in training process
In, positive sample includes the local feature of each form of the object, and negative sample is local background's feature of the object
And the feature of other related items.
6. a kind of system of target identification, which is characterized in that including detection module, the first determination module and the second determination module,
Wherein:
The detection module, for using, whether the default part containing object is special in default detection of classifier given image
Sign, and obtain testing result；
First determination module, for the testing result according to the detection module, if so, determining in the given image
Include the object；
Second determination module, for the testing result according to the detection module, if it is not, then determining in the given image
Not comprising the object；
Wherein, the default local feature is its in addition to the object of being different from for including in the selected object
The feature of the component home of his object, and the component home is a part of the object；
The quantity of the default classifier is identical as the quantity for the object to be detected, and each default detection of classifier one
The kind object；
When the detection module carries out default local feature detection to the given image using each default classifier, give
Out in the given image whether include the default local feature scoring；
It further include judging submodule in the detection module, for judging whether the scoring is greater than default score value, if so,
Obtain the testing result for "Yes"；
If it is not, then obtaining the testing result for "No".
7. the system of target identification according to claim 6, which is characterized in that the default classifier includes N layers, respectively
Classifier to n-th layer is preset for first layer and presets classifier, and the quantity of every layer of classifier is at least two；
The detection module includes N number of detection monitoring submodule, respectively the first detection sub-module to N detection sub-module, institute
State detection module when being detected,
Firstly, the first detection sub-module, which presets classifier using first layer, carries out first layer classification to the given image；
Secondly, the second detection sub-module is classified according to the first layer classification results of the first detection sub-module using the corresponding second layer
Device continues second layer classification；
And so on；
Until whether N detection sub-module contains default part using in given image described in corresponding n-th layer detection of classifier
Feature, and obtain testing result；
Wherein, N is natural number, and is more than or equal to 2.
8. the system of target identification according to claim 6, which is characterized in that further include screening in the detection module
Module, for determining the described of highest scoring when obtaining the testing result for "Yes" there are two the default classifier of the above
Default classifier obtains the testing result for "Yes", and determining remaining score not is that the highest default classifier obtains as "No"
Testing result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410720074.2A CN105095848B (en) | 2014-12-01 | 2014-12-01 | The method and system of target identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410720074.2A CN105095848B (en) | 2014-12-01 | 2014-12-01 | The method and system of target identification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105095848A CN105095848A (en) | 2015-11-25 |
| CN105095848B true CN105095848B (en) | 2019-06-21 |
Family
ID=54576239
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410720074.2A Active CN105095848B (en) | 2014-12-01 | 2014-12-01 | The method and system of target identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105095848B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106203514B (en) * | 2016-07-12 | 2019-02-12 | 腾讯科技（深圳）有限公司 | The method and apparatus of image recognition readjustment notice |
| CN108664848B (en) * | 2017-03-30 | 2020-12-25 | 杭州海康威视数字技术股份有限公司 | Image target identification method and device |
| CN108205684B (en) * | 2017-04-25 | 2022-02-11 | 北京市商汤科技开发有限公司 | Image disambiguation method, device, storage medium and electronic equipment |
| CN107145879A (en) * | 2017-06-23 | 2017-09-08 | 依通（北京）科技有限公司 | A kind of floristics automatic identifying method and system |
| CN108010068A (en) * | 2017-11-29 | 2018-05-08 | 中国人民解放军火箭军工程大学 | Ground time critical target recognition methods based on gradient direction characteristic point pair |
| CN109960990B (en) * | 2017-12-26 | 2022-06-03 | 财团法人车辆研究测试中心 | Method for evaluating reliability of obstacle detection |
| CN108491825A (en) * | 2018-03-30 | 2018-09-04 | 百度在线网络技术（北京）有限公司 | information generating method and device |
| CN109117819B (en) * | 2018-08-30 | 2021-03-02 | Oppo广东移动通信有限公司 | Target object identification method and device, storage medium and wearable device |
| CN109490904B (en) * | 2018-11-15 | 2021-09-21 | 上海炬佑智能科技有限公司 | Time-of-flight sensor and detection method thereof |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196979A (en) * | 2006-12-22 | 2008-06-11 | 四川川大智胜软件股份有限公司 | Method for recognizing vehicle type by digital picture processing technology |
| CN103824052A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Multilevel semantic feature-based face feature extraction method and recognition method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104156697A (en) * | 2014-07-24 | 2014-11-19 | 佳都新太科技股份有限公司 | Vehicle type recognition method under night bayonet scene |
-
2014
- 2014-12-01 CN CN201410720074.2A patent/CN105095848B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196979A (en) * | 2006-12-22 | 2008-06-11 | 四川川大智胜软件股份有限公司 | Method for recognizing vehicle type by digital picture processing technology |
| CN103824052A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Multilevel semantic feature-based face feature extraction method and recognition method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105095848A (en) | 2015-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105095848B (en) | The method and system of target identification | |
| CN109800824B (en) | Pipeline defect identification method based on computer vision and machine learning | |
| CN103903005B (en) | License plate image identification system and method | |
| CN105809146B (en) | A kind of image scene recognition methods and device | |
| US9684958B2 (en) | Image processing device, program, image processing method, computer-readable medium, and image processing system | |
| CN111126224A (en) | Vehicle detection method and classification recognition model training method | |
| CN103136504B (en) | Face identification method and device | |
| CN109829467A (en) | Image labeling method, electronic device and non-transient computer-readable storage medium | |
| CN105224947B (en) | classifier training method and system | |
| CN106951899A (en) | Method for detecting abnormality based on image recognition | |
| CN106022285A (en) | Vehicle type identification method and vehicle type identification device based on convolutional neural network | |
| CN107273832B (en) | License plate recognition method and system based on integral channel characteristics and convolutional neural network | |
| CN103218610B (en) | The forming method of dog face detector and dog face detecting method | |
| CN105095884A (en) | Pedestrian recognition system and pedestrian recognition processing method based on random forest support vector machine | |
| CN104615986A (en) | Method for utilizing multiple detectors to conduct pedestrian detection on video images of scene change | |
| CN111242899B (en) | Image-based flaw detection method and computer-readable storage medium | |
| CN108764302A (en) | A kind of bill images sorting technique based on color characteristic and bag of words feature | |
| CN111382766A (en) | Equipment fault detection method based on fast R-CNN | |
| WO2018067080A1 (en) | A marine vessel identification method | |
| CN103177266A (en) | Intelligent stock pest identification system | |
| CN105096293B (en) | Method and apparatus for handling the block to be processed of sediment urinalysis image | |
| EP3118812A1 (en) | Image processing device, image sensor, and image processing method | |
| Tavallali et al. | An efficient training procedure for viola-jones face detector | |
| WO2015146113A1 (en) | Identification dictionary learning system, identification dictionary learning method, and recording medium | |
| CN105160292A (en) | Vehicle identification recognition method and system |
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 |