CN110472499A - A kind of method and device that pedestrian identifies again - Google Patents
A kind of method and device that pedestrian identifies again Download PDFInfo
- Publication number
- CN110472499A CN110472499A CN201910616632.3A CN201910616632A CN110472499A CN 110472499 A CN110472499 A CN 110472499A CN 201910616632 A CN201910616632 A CN 201910616632A CN 110472499 A CN110472499 A CN 110472499A
- Authority
- CN
- China
- Prior art keywords
- sample
- threshold
- model
- preset
- threshold value
- 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
- 238000000034 method Methods 0.000 title claims abstract description 40
- 238000006243 chemical reaction Methods 0.000 claims abstract description 30
- 238000004891 communication Methods 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 9
- 238000012549 training Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 4
- 238000013507 mapping Methods 0.000 description 4
- 238000012512 characterization method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
Abstract
The invention discloses a kind of method and devices that pedestrian identifies again, are related to technical field of data processing, to increase business side's workload when solving the problems, such as that service side replaces identification model in the prior art.This method is specifically included that according to preset identification model, establishes threshold map relation table；Obtain sample to be tested, two images identified in the sample to be tested comprising needs；According to preset recognition rule, Model of Target Recognition is chosen from the preset identification model；According to the Model of Target Recognition, the distortion score of sample to be tested is calculated；According to threshold value relationship conversion formula and the threshold map relation table, the distortion score is converted into standard similarity score；Judge whether two images in the sample to be tested are same pedestrian image according to the standard similarity score.Identified present invention is mainly applied to pedestrian again during.
Description
Technical field
The present invention relates to a kind of technical field of data processing, more particularly to a kind of method and device that pedestrian identifies again.
Background technique
Pedestrian identifies the technology for just referring to and judging to whether there is specific pedestrian in image or video sequence again.The prior art
In, the feature vector to be identified of image or video sequence is calculated by identification model, then according to feature vector to be identified and spy
The similitude for determining the feature vector of pedestrian carries out pedestrian and identifies again.
It is identified in (re-id) system again in pedestrian, generally includes the back-end services system that pedestrian identifies again and (referred to as service
Side), and the front end service system (abbreviation business side) for calling service side to realize that pedestrian identifies again.It is being identified using pedestrian again
When system, business side sends two images comprising pedestrian to service side, and service orientation business side returns to a similarity score, industry
Whether business root is same people according to the pedestrian of this two images of the threshold decision of business side's setting themselves.For example, if taking
When business side uses identification model A, the value range of similarity score is [0,1], can when in business side, given threshold is 0.5
So that negative sample error rate reaches 0.01%, however after identification model is changed to Model B by service side, similarity score takes
Value range becomes [10,20], negative sample error rate could be made to reach 0.01% when given threshold is 14 in business side, this
When, if it is 0.5 that business side, which continues given threshold, the similarity score of all samples can all be greater than 0.5, can then make
At 100% negative sample error rate.
The similarity score that same two pedestrian images obtain is compared using different identification models to be different, it is similar
The value range for spending score is not also identical.If modifying model every time in order to guarantee identical comparison result accuracy and requiring
Business root resets threshold value according to modified model, increases the workload of business side.
Summary of the invention
In view of this, the present invention provides a kind of method and device that pedestrian identifies again, main purpose is to solve existing skill
The problem of increasing business side's workload in art when service side replaces identification model.
According to the present invention on one side, it provides a kind of pedestrian and knows method for distinguishing again, comprising:
According to preset identification model, threshold map relation table is established, the quantity of the preset identification model is at least a kind,
It include level threshold value, actual threshold and negative sample error rate/positive sample percent of pass in the threshold map relation table；
Obtain sample to be tested, two images identified in the sample to be tested comprising needs；
According to preset recognition rule, Model of Target Recognition is chosen from the preset identification model；
According to the Model of Target Recognition, the distortion score of sample to be tested is calculated；
According to threshold value relationship conversion formula and the threshold map relation table, the distortion score is converted into mark
Quasi- similarity score；
Judge whether two images in the sample to be tested are same pedestrian image according to the standard similarity score.
According to the present invention on the other hand, a kind of device that pedestrian identifies again is provided, comprising:
Module is established, for establishing threshold map relation table, the number of the preset identification model according to preset identification model
Amount is at least a kind, includes that level threshold value, actual threshold and negative sample error rate/positive sample are logical in the threshold map relation table
Cross rate；
Module is obtained, for obtaining sample to be tested, two images identified in the sample to be tested comprising needs；
Module is chosen, for choosing Model of Target Recognition from the preset identification model according to preset recognition rule；
Computing module, for calculating the distortion score of sample to be tested according to the Model of Target Recognition；
Conversion module is used for according to threshold value relationship conversion formula and the threshold map relation table, and the model is similar
Degree score is converted to standard similarity score；
Judgment module, for according to the standard similarity score judge two images in the sample to be tested whether be
Same pedestrian image.
According to another aspect of the invention, a kind of storage medium is provided, at least one is stored in the storage medium can
It executes instruction, the executable instruction executes processor as again above-mentioned pedestrian knows the corresponding operation of method for distinguishing.
In accordance with a further aspect of the present invention, a kind of computer equipment is provided, comprising: processor, memory, communication interface
And communication bus, the processor, the memory and the communication interface complete mutual lead to by the communication bus
Letter；
For the memory for storing an at least executable instruction, it is above-mentioned that the executable instruction executes the processor
Pedestrian knows the corresponding operation of method for distinguishing again.
By above-mentioned technical proposal, technical solution provided in an embodiment of the present invention is at least had the advantage that
The present invention provides a kind of method and devices that pedestrian identifies again, establish threshold value according to preset identification model first and reflect
Relation table is penetrated, sample to be tested is then obtained, chooses Model of Target Recognition from preset identification model further according to preset recognition rule,
The distortion score that sample to be tested is calculated further according to Model of Target Recognition, reflects further according to threshold value relationship conversion formula and threshold value
Relation table is penetrated, distortion score is converted into standard similarity score, finally according to the judgement of standard similarity score
Whether two images in sample to be tested are same pedestrian image.Compared with prior art, the embodiment of the present invention is reflected by threshold value
Relation table is penetrated, distortion score is converted into standard similarity score, so that identical standard similarity score is corresponding
Negative sample error rate/positive sample percent of pass is identical, and business side does not need to update identification model according to service side and change threshold value
Identical recognition accuracy can be reached, realize that service side updates identification model and do not increase business side's workload simultaneously.
The above description is only an overview of the technical scheme of the present invention, in order to better understand the technical means of the present invention,
And it can be implemented in accordance with the contents of the specification, and in order to allow above and other objects of the present invention, feature and advantage can
It is clearer and more comprehensible, the followings are specific embodiments of the present invention.
Detailed description of the invention
By reading the following detailed description of the preferred embodiment, various other advantages and benefits are common for this field
Technical staff will become clear.The drawings are only for the purpose of illustrating a preferred embodiment, and is not considered as to the present invention
Limitation.And throughout the drawings, the same reference numbers will be used to refer to the same parts.In the accompanying drawings:
Fig. 1 shows the method flow diagram that a kind of pedestrian provided in an embodiment of the present invention identifies again；
Fig. 2 shows the method flow diagrams that another pedestrian provided in an embodiment of the present invention identifies again；
Fig. 3 shows the device composition block diagram that a kind of pedestrian provided in an embodiment of the present invention identifies again；
Fig. 4 shows the device composition block diagram that another pedestrian provided in an embodiment of the present invention identifies again；
Fig. 5 shows a kind of structural schematic diagram of computer equipment provided in an embodiment of the present invention.
Specific embodiment
Exemplary embodiments of the present disclosure are described in more detail below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
It is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
It is fully disclosed to those skilled in the art.
It is identified in (re-id) system again in pedestrian, generally includes the back-end services system that pedestrian identifies again and (referred to as service
Side), and the front end service system (abbreviation business side) for calling service side to realize that pedestrian identifies again.It is being identified using pedestrian again
When system, business side sends two images comprising pedestrian to service side, and service orientation business side returns to a similarity score, industry
Whether business root is same people according to the pedestrian of this two images of the threshold decision of business side's setting themselves.The embodiment of the present invention provides
A kind of pedestrian knows method for distinguishing again, as shown in Figure 1, it is applied to server-side, this method comprises:
101, according to preset identification model, threshold map relation table is established.
The quantity of preset identification model is at least a kind, includes level threshold value, actual threshold and negative in threshold map relation table
Sample error rate/positive sample percent of pass.Wherein, level threshold value and negative sample error rate/positive sample percent of pass are all a column datas,
The value range of level threshold value can be [0~1], and multiple numerical value are chosen in value range and collectively form level threshold value set,
Data in level threshold value set are arranged according to according to sequence from small to large.Match for each numerical value in level threshold value set
Set negative sample error rate/positive sample percent of pass.In the columns of actual threshold in threshold map relation table and preset identification model
Quantity it is identical, in actual threshold the numerical value of every a line be using preset identification model calculate sample database in each sample it is similar
Score is spent, negative sample error rate/positive sample percent of pass actual threshold can be reached.Sample database refers to the preset identification model of test
When precision, the test set comprising m positive sample and n negative sample used, each positive sample includes two and belongs to same people's
Pedestrian image, each negative sample include two pedestrian images for being not belonging to same people.
Illustratively, the threshold map relation table on the basis of negative sample error rate includes model A in preset identification model
And Model B, it is assumed that test set S includes m positive sample and n negative sample, the threshold map relation table of foundation are as follows:
| Level threshold value | Negative sample error rate | Actual threshold (model A) | Actual threshold (Model B) |
| 0 | 100% | tA0 | tB0 |
| 0.2 | 10% | tA1 | tB1 |
| 0.3 | 1% | tA2 | tB2 |
| 0.4 | 0.1% | tA3 | tB3 |
| 0.5 | 0.01% | tA4 | tB4 |
| 0.6 | 0.001% | tA5 | tB5 |
| 0.7 | 0.0001% | tA6 | tB6 |
| 0.8 | 0.00001% | tA7 | tB7 |
| 1 | 0 | tA8 | tB8 |
For seeking tA2, a similarity score is calculated to each sample in S in model A, if given threshold
T, statistics obtain the positive sample number m0 and negative sample number n0 greater than t, calculate when the ratio of n0 and n is approximately equal to 1% value tA2.
Due to the negative sample limited amount in S, it is exactly equal to 1% so tending not to obtain in ratio calculated, so taking approximation etc.
In 1% numerical value.The required corresponding actual threshold tB0 of actual threshold tA0~tA8 and Model B is sought with similar mode
~tB8.
102, sample to be tested is obtained.
Service side receives the sample to be tested that business side uploads, two images identified in sample to be tested comprising needs.
103, according to preset recognition rule, Model of Target Recognition is chosen from the preset identification model.
Preset recognition rule, which can be, chooses corresponding Model of Target Recognition according to the identification scene that business side uploads, can also
To choose Model of Target Recognition according to sample to be tested.Identification scene may be identification criminal, identify that the old man that wanders away, identification wander away
Child etc., for different identification scenes, if identification criminal, it is contemplated that criminal may be changed during escaping and wear
, increase the camouflage such as accessories, so Model of Target Recognition need using posture, face etc. not malleable feature as main identification according to
According to a possibility that reduce omission；If identification is wandered away old man, it is contemplated that old man during wandering away, clothes changing accessories
Possibility is smaller, so Model of Target Recognition is screened with clothing color, again to features such as image posture, the faces filtered out
Screening is done, to guarantee screening speed.What it is due to sample to be tested may be certificate photo, living photo or video interception, then to sample
In two images may be two distinct types of photo, then the pedestrian that can be extracted in two images is characterized in difference
, so can only be using common pedestrian's feature that two images can extract as basis of characterization.
Before choosing Model of Target Recognition, the mapping relations of identification scene and preset identification model are set in service side
The mapping table of table or basis of characterization and preset identification model.Then the identification scene or to be measured uploaded according to business side
The basis of characterization of sample chooses Model of Target Recognition.
104, according to the Model of Target Recognition, the distortion score of sample to be tested is calculated.
The distortion score of sample to be tested refers to the similarity for calculating two kinds of images in sample to be tested.Calculating phase
When like degree score, is calculated according to Model of Target Recognition, Histogram Matching, matrix decomposition can be used in Model of Target Recognition
Or the method based on characteristic point compares the image in sample to be tested.
105, according to threshold value relationship conversion formula and the threshold map relation table, the distortion score is converted
For standard similarity score.
Threshold value relationship conversion formula, for distortion score to be converted to standard similarity score, foundation is threshold
It is worth mapping table.In calculating process, it is first determined the corresponding actual threshold of Model of Target Recognition in threshold map relation table,
Then actual threshold range locating for similarity score, the reality in actual threshold range i.e. threshold map relation table are determined
With immediate 2 data of distortion score in threshold value, then level threshold value range corresponding with actual threshold range is searched,
Similar with actual threshold range, level threshold value range is 2 adjacent numerical value, finally according to level threshold value range, actual threshold
Range and distortion score calculate standard similarity score according to threshold value relationship conversion formula.
106, judge whether two images in the sample to be tested are same a group traveling together according to the standard similarity score
Image.
After business side receives standard similarity score, if standard similarity score is greater than the given threshold of user,
Judge two images in sample to be tested for same pedestrian image；If standard similarity score is not more than the setting threshold of user
Value, then judge that two in sample to be tested images are not same pedestrian image.Given threshold is equivalent in threshold map relation table
Level threshold value, when given threshold is constant, even if the Model of Target Recognition that uses of service side is different, the judgement knot of sample to be tested
Negative sample error rate/positive sample percent of pass of fruit also remains unchanged.
The present invention provides a kind of pedestrians to know method for distinguishing again, establishes threshold map relationship according to preset identification model first
Then table obtains sample to be tested, choose Model of Target Recognition from preset identification model further according to preset recognition rule, further according to
Model of Target Recognition calculates the distortion score of sample to be tested, further according to threshold value relationship conversion formula and threshold map relationship
Distortion score is converted to standard similarity score by table, finally described to test sample according to the judgement of standard similarity score
Whether two images in this are same pedestrian image.Compared with prior art, the embodiment of the present invention passes through threshold map relationship
Distortion score is converted to standard similarity score by table, so that the corresponding negative sample of identical standard similarity score
Error rate/positive sample percent of pass is identical, and business side does not need to update identification model according to service side and changing threshold value can also reach
To identical recognition accuracy, realize that service side updates identification model and do not increase business side's workload simultaneously.
The embodiment of the invention provides another pedestrians to know method for distinguishing again, as shown in Fig. 2, this method comprises:
201, according to the identification scene, preset bulk sample is originally divided at least one subsample, the subsample and institute
Identification scene is stated to correspond.
Identification scene includes identification criminal, identifies the old man that wanders away, identification lost children etc., for different identification fields
Preset bulk sample is originally divided into subsample corresponding with identification scene by scape, for example runaway convict and the old man that wanders away belong to different increments
This.
202, according to the subsample, the identical preset identification model of training identification scene corresponding with the subsample.
The corresponding identification scene in subsample is searched, and corresponding preset identification model is searched according to identification scene, it will be different
Preset identification model, be trained using with the corresponding subsample of identification scene.Illustratively, with runaway convict subsample training mould
Type A, then model A would be even more beneficial to distinguish different runaway convicts, implementation model A and the identification Scene realization of identification runaway convict are real
Corresponding relationship, similarly wander away old man subsample training pattern B, then Model B would be even more beneficial to distinguish different old men.It is instructing
During white silk, model A and Model B can accomplish real-time training in lesser calculating cost, so that the real-time of training
And it calculates and reaches balance between cost.
203, according to preset identification model, threshold map relation table is established.
The quantity of preset identification model is at least a kind, includes level threshold value, actual threshold in the threshold map relation table
With negative sample error rate/positive sample percent of pass.Specific establishment process includes: in preset codomain according to preset step size computation institute
State level threshold value；Configure negative sample error rate/positive sample percent of pass corresponding with the level threshold value；According to preset sample database,
It calculates the preset identification model and reaches the negative sample error rate/corresponding actual threshold of positive sample percent of pass；According to
The level threshold value, the actual threshold and the negative sample error rate/positive sample percent of pass, generate the threshold map
Relation table.
The value range of preset codomain can be [0~1], and preset step-length can choose 0.1, according to preset in preset codomain
Step-length chooses multiple numerical value and collectively forms level threshold value set, by the data in level threshold value set according to according to from small to large
Sequence arranges.Illustratively, the threshold map relation table on the basis of positive sample percent of pass includes model in preset identification model
A and Model B, it is assumed that test set S includes m positive sample and n negative sample, the threshold map relation table of foundation are as follows:
| Level threshold value | Positive sample percent of pass | Actual threshold (model A) | Actual threshold (Model B) |
| 0 | 100% | tA0 | tB0 |
| 0.1 | 90% | tA1 | tB1 |
| 0.2 | 80% | tA2 | tB2 |
| 0.3 | 70% | tA3 | tB3 |
| 0.4 | 60% | tA4 | tB4 |
| 0.5 | 50% | tA5 | tB5 |
| 0.6 | 40% | tA6 | tB6 |
| 0.7 | 30% | tA7 | tB7 |
| 0.8 | 20% | tA8 | tB8 |
| 0.9 | 10% | tA9 | tB9 |
| 1 | 0 | tA10 | tB10 |
For seeking tA2, a similarity score is calculated to each sample in S in model A, if given threshold
T, statistics obtain the positive sample number m0 and negative sample number n0 greater than t, calculate when the ratio of m0 and m is approximately equal to 80% value tA2.
Due to the negative sample limited amount in S, it is exactly equal to 80% so tending not to obtain in ratio calculated, so taking approximation etc.
In 80% numerical value.Required actual threshold tA0~tA8 and the corresponding actual threshold of Model B are sought with similar mode
TB0~tB8.
Before establishing threshold map relation table, it is also necessary to judge whether business side was once gone using a certain identification model
People identifies again, if it is judged that be, using the actual threshold of the last identification model used as preset mapping relations
Level threshold value in table, so that business side user does not need to reset threshold value according to new identification model, with utmostly
Reduction business side user workload.If it is judged that be it is no, then by facilitate calculate for the purpose of be arranged threshold map relationship
Level threshold value in table.
In view of pedestrian's weight identification model is continuously improved in use, and new identification mould may also be invented
Type, so after establishing threshold map relation table improved identification model or new identification model may be used in service side.
For the situation, the more new channel of preset identification model is set in service side, if receiving the preset identification model more
New command calculates the corresponding update identification model of more new command and reaches the negative sample mistake then according to preset sample database
The corresponding update actual threshold of rate/positive sample percent of pass；The update actual threshold is saved to the threshold map and is closed
It is table.
When updating threshold map relation table, the corresponding actual threshold of the identification model abandoned can also be deleted, if institute
There is business side currently in use whether all to abandon the identification model and delete this again and abandons the corresponding actual threshold of model.
204, sample to be tested is obtained.
Service side receives the sample to be tested that business side uploads, two images identified in sample to be tested comprising needs.
205, according to preset recognition rule, Model of Target Recognition is chosen from the preset identification model.
Preset recognition rule, which can be, chooses corresponding Model of Target Recognition according to the identification scene that business side uploads, can also
To choose Model of Target Recognition according to sample to be tested.For choosing Model of Target Recognition according to identification scene, specifically includes: obtaining
The scene of the preset identification model and identification scene is applicable in the table of comparisons；It is applicable in the table of comparisons in the scene, searches and preselect
Identify the corresponding Model of Target Recognition of scene.Pre-selection identification scene can be uploaded by business side, can also be according to sample to be tested
Sample characteristics are chosen.
206, according to the Model of Target Recognition, the distortion score of sample to be tested is calculated.
Euclidean distance and cosine similarity are all the measures of vector distance, and Euclidean distance should consider the direction of vector
The length of vector is considered again, and cosine similarity only considers direction, it is small using the calculation amount of cosine similarity, so of the invention
Distortion score be cosine similarity between two images in sample to be tested.
207, according to threshold value relationship conversion formula and the threshold map relation table, the distortion score is converted
For standard similarity score.
The conversion method of standard similarity score specifically includes: in the threshold map relation table, searching the target
Actual threshold corresponding to identification model；Extract distortion score reality corresponding to the Model of Target Recognition
Actual threshold range belonging in threshold value；Extract the corresponding level threshold value range of the actual threshold range；According to the threshold value
Relationship conversion formula, the level threshold value range and the actual threshold range, it is corresponding to calculate the distortion score
Standard similarity score, the threshold value relationship conversion formula areWherein s' is the standard phase
Like degree score, s is the distortion score, t-Under in actual threshold range belonging to the distortion score
Boundary, t+For the coboundary in actual threshold range belonging to the distortion score, s-For with t-Corresponding level threshold value
Lower boundary in range, s+For with t+Coboundary in corresponding level threshold value range.
Illustratively, it is assumed that the actual threshold range of distortion score s is tA3~tA4, corresponding level threshold value model
0.3~0.4 is enclosed, s is substituted into formula by level threshold value range,Calculating and distortion
The corresponding standard similarity score of score.
208, judge whether two images in the sample to be tested are same a group traveling together according to the standard similarity score
Image.
After business side receives standard similarity score, if standard similarity score is greater than the given threshold of user,
Judge two images in sample to be tested for same pedestrian image；If standard similarity score is not more than the setting threshold of user
Value, then judge that two in sample to be tested images are not same pedestrian image.Given threshold is equivalent in threshold map relation table
Level threshold value, when given threshold is constant, even if the Model of Target Recognition that uses of service side is different, the judgement knot of sample to be tested
Negative sample error rate/positive sample percent of pass of fruit also remains unchanged.
The present invention provides a kind of pedestrians to know method for distinguishing again, establishes threshold map relationship according to preset identification model first
Then table obtains sample to be tested, choose Model of Target Recognition from preset identification model further according to preset recognition rule, further according to
Model of Target Recognition calculates the distortion score of sample to be tested, further according to threshold value relationship conversion formula and threshold map relationship
Distortion score is converted to standard similarity score by table, finally described to test sample according to the judgement of standard similarity score
Whether two images in this are same pedestrian image.Compared with prior art, the embodiment of the present invention passes through threshold map relationship
Distortion score is converted to standard similarity score by table, so that the corresponding negative sample of identical standard similarity score
Error rate/positive sample percent of pass is identical, and business side does not need to update identification model according to service side and changing threshold value can also reach
To identical recognition accuracy, realize that service side updates identification model and do not increase business side's workload simultaneously.
Further, as the realization to method shown in above-mentioned Fig. 1, the embodiment of the invention provides a kind of pedestrians to identify again
Device, as shown in figure 3, the device includes:
Module 31 is established, for establishing threshold map relation table according to preset identification model, the preset identification model
Quantity is at least a kind, includes level threshold value, actual threshold and negative sample error rate/positive sample in the threshold map relation table
Percent of pass；
Module 32 is obtained, for obtaining sample to be tested, two images identified in the sample to be tested comprising needs；
Module 33 is chosen, for choosing Model of Target Recognition from the preset identification model according to preset recognition rule；
Computing module 34, for calculating the distortion score of sample to be tested according to the Model of Target Recognition；
Conversion module 35 is used for according to threshold value relationship conversion formula and the threshold map relation table, by the model phase
Standard similarity score is converted to like degree score；
Judgment module 36, for whether judging two images in the sample to be tested according to the standard similarity score
For same pedestrian image.
The present invention provides a kind of devices that pedestrian identifies again, establish threshold map relationship according to preset identification model first
Then table obtains sample to be tested, choose Model of Target Recognition from preset identification model further according to preset recognition rule, further according to
Model of Target Recognition calculates the distortion score of sample to be tested, further according to threshold value relationship conversion formula and threshold map relationship
Distortion score is converted to standard similarity score by table, finally described to test sample according to the judgement of standard similarity score
Whether two images in this are same pedestrian image.Compared with prior art, the embodiment of the present invention passes through threshold map relationship
Distortion score is converted to standard similarity score by table, so that the corresponding negative sample of identical standard similarity score
Error rate/positive sample percent of pass is identical, and business side does not need to update identification model according to service side and changing threshold value can also reach
To identical recognition accuracy, realize that service side updates identification model and do not increase business side's workload simultaneously.
Further, as the realization to method shown in above-mentioned Fig. 2, the embodiment of the invention provides another pedestrians to know again
Other device, as shown in figure 4, the device includes:
Module 41 is established, for establishing threshold map relation table according to preset identification model, the preset identification model
Quantity is at least a kind, includes level threshold value, actual threshold and negative sample error rate/positive sample in the threshold map relation table
Percent of pass；
Module 42 is obtained, for obtaining sample to be tested, two images identified in the sample to be tested comprising needs；
Module 43 is chosen, for choosing Model of Target Recognition from the preset identification model according to preset recognition rule；
Computing module 44, for calculating the distortion score of sample to be tested according to the Model of Target Recognition；
Conversion module 45 is used for according to threshold value relationship conversion formula and the threshold map relation table, by the model phase
Standard similarity score is converted to like degree score；
Judgment module 46, for whether judging two images in the sample to be tested according to the standard similarity score
For same pedestrian image.
It is further, described to establish module 41, comprising:
First computing unit 411, for the level threshold value according to preset step size computation in preset codomain；
Configuration unit 412, for configuring negative sample error rate/positive sample percent of pass corresponding with the level threshold value；
Second computing unit 413 is also used to calculate the preset identification model according to preset sample database and reach the negative sample
This error rate/corresponding actual threshold of positive sample percent of pass；
Generation unit 414, for according to the level threshold value, the actual threshold and the negative sample error rate/described
Positive sample percent of pass generates the threshold map relation table.
Further, the selection module 43, comprising:
Acquiring unit 431, for obtaining the preset identification model and identifying that the scene of scene is applicable in the table of comparisons；
Searching unit 432 is searched target corresponding with pre-selection identification scene and is known for being applicable in the table of comparisons in the scene
Other model.
Further, the method also includes:
Division module 47 is established before threshold map relation table for described according to preset identification model according to the knowledge
Preset bulk sample is originally divided at least one subsample by other scene, and the subsample and the identification scene correspond；
Training module 48, for according to the subsample, training identification scene corresponding with the subsample to be identical pre-
Set identification model.
Further, the conversion module 45, comprising:
Searching unit 451, for searching reality corresponding to the Model of Target Recognition in the threshold map relation table
Border threshold value；
Extraction unit 452, for extracting distortion score reality corresponding to the Model of Target Recognition
Actual threshold range belonging in threshold value；
The extraction unit 452 is also used to extract the corresponding level threshold value range of the actual threshold range；
Computing unit 453, for according to the threshold value relationship conversion formula, the level threshold value range and the practical threshold
It is worth range, calculates the corresponding standard similarity score of the distortion score, the threshold value relationship conversion formula isWherein s' is the standard similarity score, and s is the distortion score, t-For institute
State the lower boundary in actual threshold range belonging to distortion score, t+For reality belonging to the distortion score
Coboundary in threshold range, s-For with t-Lower boundary in corresponding level threshold value range, s+For with t+Corresponding level threshold value
Coboundary in range.
Further, the method also includes:
Update module 49 is established after threshold map relation table for described according to preset identification model, if received
The more new command of the preset identification model calculates the corresponding update of the more new command and identifies mould then according to preset sample database
Type reaches the corresponding update actual threshold of the negative sample error rate/positive sample percent of pass；
Preserving module 410, for saving the update actual threshold to the threshold map relation table.
Further, the distortion score is the cosine similarity in the sample to be tested between two images.
The present invention provides a kind of devices that pedestrian identifies again, establish threshold map relationship according to preset identification model first
Then table obtains sample to be tested, choose Model of Target Recognition from preset identification model further according to preset recognition rule, further according to
Model of Target Recognition calculates the distortion score of sample to be tested, further according to threshold value relationship conversion formula and threshold map relationship
Distortion score is converted to standard similarity score by table, finally described to test sample according to the judgement of standard similarity score
Whether two images in this are same pedestrian image.Compared with prior art, the embodiment of the present invention passes through threshold map relationship
Distortion score is converted to standard similarity score by table, so that the corresponding negative sample of identical standard similarity score
Error rate/positive sample percent of pass is identical, and business side does not need to update identification model according to service side and changing threshold value can also reach
To identical recognition accuracy, realize that service side updates identification model and do not increase business side's workload simultaneously.
A kind of storage medium is provided according to an embodiment of the present invention, and it is executable that the storage medium is stored at least one
Instruction, the pedestrian which can be performed in above-mentioned any means embodiment know method for distinguishing again.
Fig. 5 shows a kind of structural schematic diagram of the computer equipment provided according to an embodiment of the present invention, the present invention
Specific embodiment does not limit the specific implementation of computer equipment.
As shown in figure 5, the computer equipment may include: processor (processor) 502, communication interface
(Communications Interface) 504, memory (memory) 506 and communication bus 508.
Wherein: processor 502, communication interface 504 and memory 506 complete mutual lead to by communication bus 508
Letter.
Communication interface 504, for being communicated with the network element of other equipment such as client or other servers etc..
Processor 502 can specifically execute in the embodiment of the method that above-mentioned pedestrian identifies again for executing program 510
Correlation step.
Specifically, program 510 may include program code, which includes computer operation instruction.
Processor 502 may be central processor CPU or specific integrated circuit ASIC (Application
Specific Integrated Circuit), or be arranged to implement the integrated electricity of one or more of the embodiment of the present invention
Road.The one or more processors that computer equipment includes can be same type of processor, such as one or more CPU；
It can be different types of processor, such as one or more CPU and one or more ASIC.
Memory 506, for storing program 510.Memory 506 may include high speed RAM memory, it is also possible to further include
Nonvolatile memory (non-volatile memory), for example, at least a magnetic disk storage.
Program 510 specifically can be used for so that processor 502 executes following operation:
According to preset identification model, threshold map relation table is established, the quantity of the preset identification model is at least a kind,
It include level threshold value, actual threshold and negative sample error rate/positive sample percent of pass in the threshold map relation table；
Obtain sample to be tested, two images identified in the sample to be tested comprising needs；
According to preset recognition rule, Model of Target Recognition is chosen from the preset identification model；
According to the Model of Target Recognition, the distortion score of sample to be tested is calculated；
According to threshold value relationship conversion formula and the threshold map relation table, the distortion score is converted into mark
Quasi- similarity score；
Judge whether two images in the sample to be tested are same pedestrian image according to the standard similarity score.
Obviously, those skilled in the art should be understood that each module of the above invention or each step can be with general
Computing device realize that they can be concentrated on a single computing device, or be distributed in multiple computing devices and formed
Network on, optionally, they can be realized with the program code that computing device can perform, it is thus possible to which they are stored
It is performed by computing device in the storage device, and in some cases, it can be to be different from shown in sequence execution herein
Out or description the step of, perhaps they are fabricated to each integrated circuit modules or by them multiple modules or
Step is fabricated to single integrated circuit module to realize.In this way, the present invention is not limited to any specific hardware and softwares to combine.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all include within protection scope of the present invention.
Claims (10)
1. a kind of pedestrian knows method for distinguishing again characterized by comprising
According to preset identification model, threshold map relation table is established, the quantity of the preset identification model is at least a kind, described
It include level threshold value, actual threshold and negative sample error rate/positive sample percent of pass in threshold map relation table；
Obtain sample to be tested, two images identified in the sample to be tested comprising needs；
According to preset recognition rule, Model of Target Recognition is chosen from the preset identification model；
According to the Model of Target Recognition, the distortion score of sample to be tested is calculated；
According to threshold value relationship conversion formula and the threshold map relation table, the distortion score is converted into standard phase
Like degree score；
Judge whether two images in the sample to be tested are same pedestrian image according to the standard similarity score.
2. the method as described in claim 1, which is characterized in that it is described according to preset identification model, establish threshold map relationship
Table, comprising:
The level threshold value according to preset step size computation in preset codomain；
Configure negative sample error rate/positive sample percent of pass corresponding with the level threshold value；
According to preset sample database, calculates the preset identification model and reach the negative sample error rate/positive sample percent of pass
Corresponding actual threshold；
According to the level threshold value, the actual threshold and the negative sample error rate/positive sample percent of pass, described in generation
Threshold map relation table.
3. the method as described in claim 1, which is characterized in that it is described according to preset recognition rule, from the preset identification mould
Model of Target Recognition is chosen in type, comprising:
It obtains the preset identification model and identifies that the scene of scene is applicable in the table of comparisons；
It is applicable in the table of comparisons in the scene, searches Model of Target Recognition corresponding with pre-selection identification scene.
4. method as claimed in claim 3, which is characterized in that it is described according to preset identification model, establish threshold map relationship
Before table, the method also includes:
According to the identification scene, preset bulk sample is originally divided at least one subsample, the subsample and the identification field
Scape corresponds；
According to the subsample, the identical preset identification model of training identification scene corresponding with the subsample.
5. the method as described in right wants 2, which is characterized in that described to be closed according to threshold value relationship conversion formula and the threshold map
It is table, the distortion score is converted into standard similarity score, comprising:
In the threshold map relation table, actual threshold corresponding to the Model of Target Recognition is searched；
Actual threshold belonging to extracting in distortion score actual threshold corresponding to the Model of Target Recognition
Range；
Extract the corresponding level threshold value range of the actual threshold range；
According to the threshold value relationship conversion formula, the level threshold value range and the actual threshold range, the model is calculated
The corresponding standard similarity score of similarity score, the threshold value relationship conversion formula areIts
Middle s' is the standard similarity score, and s is the distortion score, t-For reality belonging to the distortion score
Lower boundary in the threshold range of border, t+For the coboundary in actual threshold range belonging to the distortion score, s-For with
t-Lower boundary in corresponding level threshold value range, s+For with t+Coboundary in corresponding level threshold value range.
6. method according to claim 2, which is characterized in that it is described according to preset identification model, establish threshold map relationship
After table, the method also includes:
If receiving the more new command of the preset identification model, according to preset sample database, the more new command pair is calculated
The update identification model answered reaches the corresponding update actual threshold of the negative sample error rate/positive sample percent of pass；
The update actual threshold is saved to the threshold map relation table.
7. as the method according to claim 1 to 6, which is characterized in that the distortion score is described to test sample
Cosine similarity in this between two images.
8. a kind of device that pedestrian identifies again characterized by comprising
Module is established, for establishing threshold map relation table, the quantity of the preset identification model is extremely according to preset identification model
Less it is a kind, includes level threshold value, actual threshold and negative sample error rate/positive sample percent of pass in the threshold map relation table；
Module is obtained, for obtaining sample to be tested, two images identified in the sample to be tested comprising needs；
Module is chosen, for choosing Model of Target Recognition from the preset identification model according to preset recognition rule；
Computing module, for calculating the distortion score of sample to be tested according to the Model of Target Recognition；
Conversion module, for according to threshold value relationship conversion formula and the threshold map relation table, the distortion to be divided
Number is converted to standard similarity score；
Judgment module, for judging whether two images in the sample to be tested are same according to the standard similarity score
Pedestrian image.
9. a kind of storage medium, it is stored with an at least executable instruction in the storage medium, the executable instruction makes to handle
Device is executed as again pedestrian of any of claims 1-7 knows the corresponding operation of method for distinguishing.
10. a kind of computer equipment, comprising: processor, memory, communication interface and communication bus, the processor described are deposited
Reservoir and the communication interface complete mutual communication by the communication bus；
The memory executes the processor as right is wanted for storing an at least executable instruction, the executable instruction
Pedestrian described in any one of 1-7 is asked to know the corresponding operation of method for distinguishing again.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910616632.3A CN110472499B (en) | 2019-07-09 | 2019-07-09 | Pedestrian re-identification method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910616632.3A CN110472499B (en) | 2019-07-09 | 2019-07-09 | Pedestrian re-identification method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110472499A true CN110472499A (en) | 2019-11-19 |
| CN110472499B CN110472499B (en) | 2024-01-30 |
Family
ID=68507578
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910616632.3A Active CN110472499B (en) | 2019-07-09 | 2019-07-09 | Pedestrian re-identification method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110472499B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582305A (en) * | 2020-03-26 | 2020-08-25 | 平安科技（深圳）有限公司 | Biological feature recognition method and device, computer equipment and storage medium |
| CN113435325A (en) * | 2021-06-25 | 2021-09-24 | 平安科技（深圳）有限公司 | Image re-recognition method and device, electronic equipment and storage medium |
| CN113469015A (en) * | 2021-06-29 | 2021-10-01 | 浙江大华技术股份有限公司 | Face recognition method and device, electronic equipment and computer storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090319513A1 (en) * | 2006-08-03 | 2009-12-24 | Nec Corporation | Similarity calculation device and information search device |
| CN107577687A (en) * | 2016-07-20 | 2018-01-12 | 北京陌上花科技有限公司 | Image search method and device |
| CN108241868A (en) * | 2016-12-26 | 2018-07-03 | 浙江宇视科技有限公司 | The objective similarity of image is to the mapping method and device of subjective similarity |
| CN108304435A (en) * | 2017-09-08 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Information recommendation method, device, computer equipment and storage medium |
| CN108804578A (en) * | 2018-05-24 | 2018-11-13 | 南京理工大学 | The unsupervised video summarization method generated based on consistency segment |
| CN109740413A (en) * | 2018-11-14 | 2019-05-10 | 平安科技（深圳）有限公司 | Pedestrian recognition methods, device, computer equipment and computer storage medium again |
-
2019
- 2019-07-09 CN CN201910616632.3A patent/CN110472499B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20090319513A1 (en) * | 2006-08-03 | 2009-12-24 | Nec Corporation | Similarity calculation device and information search device |
| CN107577687A (en) * | 2016-07-20 | 2018-01-12 | 北京陌上花科技有限公司 | Image search method and device |
| CN108241868A (en) * | 2016-12-26 | 2018-07-03 | 浙江宇视科技有限公司 | The objective similarity of image is to the mapping method and device of subjective similarity |
| CN108304435A (en) * | 2017-09-08 | 2018-07-20 | 腾讯科技（深圳）有限公司 | Information recommendation method, device, computer equipment and storage medium |
| CN108804578A (en) * | 2018-05-24 | 2018-11-13 | 南京理工大学 | The unsupervised video summarization method generated based on consistency segment |
| CN109740413A (en) * | 2018-11-14 | 2019-05-10 | 平安科技（深圳）有限公司 | Pedestrian recognition methods, device, computer equipment and computer storage medium again |
Non-Patent Citations (1)
| Title |
|---|
| 吴剑章;朱林立;高炜;: "本体算法中相似度矩阵的学习", 小型微型计算机系统, vol. 36, no. 04, pages 774 - 777 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582305A (en) * | 2020-03-26 | 2020-08-25 | 平安科技（深圳）有限公司 | Biological feature recognition method and device, computer equipment and storage medium |
| CN111582305B (en) * | 2020-03-26 | 2023-08-18 | 平安科技（深圳）有限公司 | Biological feature recognition method, apparatus, computer device and storage medium |
| CN113435325A (en) * | 2021-06-25 | 2021-09-24 | 平安科技（深圳）有限公司 | Image re-recognition method and device, electronic equipment and storage medium |
| CN113469015A (en) * | 2021-06-29 | 2021-10-01 | 浙江大华技术股份有限公司 | Face recognition method and device, electronic equipment and computer storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110472499B (en) | 2024-01-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111179311B (en) | Multi-target tracking method and device and electronic equipment | |
| CN110472499A (en) | A kind of method and device that pedestrian identifies again | |
| CN107657249A (en) | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again | |
| CN109658454B (en) | Pose information determination method, related device and storage medium | |
| CN111898739B (en) | Data screening model construction method, data screening method, device, computer equipment and storage medium based on meta learning | |
| CN110019889A (en) | Training characteristics extract model and calculate the method and relevant apparatus of picture and query word relative coefficient | |
| WO2019080908A1 (en) | Image processing method and apparatus for implementing image recognition, and electronic device | |
| US11922658B2 (en) | Pose tracking method, pose tracking device and electronic device | |
| CN106650965A (en) | Remote video processing method and apparatus | |
| CN109858552A (en) | A kind of object detection method and equipment for fine grit classification | |
| CN109074757A (en) | Method, terminal and computer readable storage medium for establishing map | |
| CN110827312A (en) | Learning method based on cooperative visual attention neural network | |
| CN109308516A (en) | A kind of method and apparatus of image procossing | |
| CN110245609A (en) | Pedestrian track generation method, device and readable storage medium storing program for executing | |
| Zhang et al. | Bing++: A fast high quality object proposal generator at 100fps | |
| CN108681708A (en) | A kind of vena metacarpea image-recognizing method, device and storage medium based on Inception neural network models | |
| CN110334743A (en) | A kind of progressive transfer learning method based on the long memory network in short-term of convolution | |
| CN112200862B (en) | Training method of target detection model, target detection method and device | |
| TWI602434B (en) | Photographing system for long-distance running event and operation method thereof | |
| CN112508135B (en) | Model training method, pedestrian attribute prediction method, device and equipment | |
| CN112052883B (en) | Clothes detection method, device and storage medium | |
| CN111866468B (en) | Object tracking distribution method, device, storage medium and electronic device | |
| CN115272923A (en) | Intelligent identification method and system based on big data platform | |
| CN111932584B (en) | Method and device for determining moving object in image | |
| Masoumzadeh et al. | HydraPicker: Fully Automated Particle Picking in Cryo-EM by Utilizing Dataset Bias in Single Shot Detection. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |