CN106971178A - Pedestrian detection and the method and device recognized again - Google Patents
Pedestrian detection and the method and device recognized again Download PDFInfo
- Publication number
- CN106971178A CN106971178A CN201710330307.1A CN201710330307A CN106971178A CN 106971178 A CN106971178 A CN 106971178A CN 201710330307 A CN201710330307 A CN 201710330307A CN 106971178 A CN106971178 A CN 106971178A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- characteristic information
- vector
- module
- subregion
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/30—Scenes; Scene-specific elements in albums, collections or shared content, e.g. social network photos or video
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The embodiments of the invention provide a kind of pedestrian detection and the method and device recognized again, this method includes：Extract the characteristic tensor of original image；According to the characteristic tensor, at least one subregion is determined；Calculate and at least one one-to-one vector characteristics of at least one described subregion；Based at least one described vector characteristics, determine the position of pedestrian in the original image and extract pedestrian's characteristic information to be identified for recognizing again.As can be seen here, the embodiment of the present invention can provide the characteristic information for ReID while pedestrian detection is carried out.The efficiency of processing can not only be so improved, and can avoid introducing extra error, it is ensured that the precision of processing.
Description
Technical field
The present invention relates to field of image recognition, relate more specifically to a kind of pedestrian detection and the method and device recognized again.
Background technology
Pedestrian detection can apply to the association areas such as intelligent driving, auxiliary driving and intelligent monitoring, mainly for detection of
It whether there is pedestrian in image or video.Pedestrian recognizes that (re-identification, ReID) is also referred to as pedestrian and recognized again again,
The association areas such as safety-security area, criminal investigation field are can apply to, are mainly used in finding with target most in the image of one group of pedestrian
One image of picture.In existing method, pedestrian detection and ReID are often considered as two separate links.In pedestrian
In detection, pedestrian is found for each two field picture, and their position and size are showed by frame.Then exist
ReID links compare the similarity of these pedestrians and target, so as to reach ReID purpose.However, due in two above link
In each link may introduce extra error, therefore by easily causing error after above-mentioned two independent links
Propagation expand, so as to cause computational accuracy low.
The content of the invention
The present invention is proposed in view of above mentioned problem.The invention provides a kind of pedestrian detection and know again method for distinguishing and
Device, can determine the characteristic information for ReID while pedestrian detection, it is to avoid introduce extra error, so as to ensure meter
The precision of calculation.
Know according to the first aspect of the invention there is provided a kind of pedestrian detection and again method for distinguishing, including：
Extract the characteristic tensor of original image；
According to the characteristic tensor, at least one subregion is determined；
Calculate and at least one one-to-one vector characteristics of at least one described subregion；
Based at least one described vector characteristics, determine in the original image position of pedestrian and extract for knowing again
Other pedestrian's characteristic information to be identified.
Exemplarily, it is described according to the characteristic tensor, at least one subregion is determined, including：
According to the characteristic tensor, multiple super-pixel points are built, each super-pixel point represents a C dimensional vector；
According to the multiple super-pixel point, it is determined that with the one-to-one multiple regions of the multiple super-pixel point；
At least one subregion according to being determined the multiple region.
Exemplarily, described at least one subregion according to being determined the multiple region, including：Using non-maximum
Suppress NMS algorithms, multiple rectangular areas are determined based on the multiple region；The corresponding feature in the multiple rectangular area is carried out
It is comprehensive, obtain at least one described subregion.
Exemplarily, the calculating and at least one one-to-one vector characteristics of at least one described subregion, including：
For every sub-regions at least one described subregion：
Each channel in the corresponding vector of all super-pixel points in every sub-regions is taken into maximum or average
Value, is obtained and the vector characteristics corresponding per sub-regions.
Exemplarily, it is described based at least one described vector characteristics, determine the position of pedestrian in the original image with
And pedestrian's characteristic information to be identified for recognizing again is extracted, including：
Based at least one described vector characteristics, the position of pedestrian in the original image is determined, wherein, the position table
It is shown as coordinate of the pedestrian in the original image；And
Based at least one described vector characteristics, pedestrian's characteristic information to be identified for recognizing again is extracted.
Exemplarily, in addition to：Based at least one described vector characteristics, the thing at least one described subregion is judged
Whether body is pedestrian.
Exemplarily, in addition to：Pedestrian's characteristic information to be identified for recognizing again is believed with target pedestrian feature
Breath is compared, to judge that the corresponding pedestrian of pedestrian's characteristic information to be identified for being used to recognize again is with the target pedestrian
No is same people.
Exemplarily, the target pedestrian characteristic information is obtained by following steps：
Extract the characteristic tensor of target pedestrian image；
According to the characteristic tensor of the target pedestrian image, target pedestrian area is determined；
At least one object vector feature corresponding with the target pedestrian area is calculated, based at least one described target
Vector characteristics, obtain the target pedestrian characteristic information.
According to the second aspect of the invention there is provided a kind of pedestrian detection and the device recognized again, including：
Extraction module, the characteristic tensor for extracting original image；
First determining module, for according to the characteristic tensor, determining at least one subregion；
Computing module, for calculating and at least one one-to-one vector characteristics of at least one described subregion；
Second determining module, for based at least one described vector characteristics, determining the position of pedestrian in the original image
Put and extract pedestrian's characteristic information to be identified for recognizing again.
Exemplarily, first determining module, including：
Submodule is built, for according to the characteristic tensor, building multiple super-pixel points, each super-pixel point represents one
C dimensional vectors；
First determination sub-module, for according to the multiple super-pixel point, it is determined that with a pair of the multiple super-pixel point 1
The multiple regions answered；
Second determination sub-module, at least one subregion according to the determination of the multiple region.
Exemplarily, second determination sub-module, is used for：Using non-maxima suppression NMS algorithms, based on the multiple
Region determines multiple rectangular areas；The corresponding feature in the multiple rectangular area is integrated, at least one described son is obtained
Region.
Exemplarily, the computing module, is used for：
For every sub-regions at least one described subregion：
Each channel in the corresponding vector of all super-pixel points in every sub-regions is taken into maximum or average
Value, is obtained and the vector characteristics corresponding per sub-regions.
Exemplarily, second determining module, is used for：
Based at least one described vector characteristics, the position of pedestrian in the original image is determined, wherein, the position table
It is shown as coordinate of the pedestrian in the original image；And
Based at least one described vector characteristics, pedestrian's characteristic information to be identified for recognizing again is extracted.
Exemplarily, second determining module, is additionally operable to：Based at least one described vector characteristics, judge described in extremely
Whether the object in few sub-regions is pedestrian.
Exemplarily, in addition to judge module, it is used for：By the pedestrian's characteristic information to be identified and mesh for recognizing again
Mark pedestrian's characteristic information is compared, to judge the corresponding pedestrian of pedestrian's characteristic information to be identified for being used to recognize again and institute
Whether state target pedestrian is same people.
Exemplarily, in addition to acquisition module, for the target pedestrian characteristic information to be obtained ahead of time；
Wherein, the acquisition module includes：
Extracting sub-module, the characteristic tensor for extracting target pedestrian image；
Determination sub-module, for the characteristic tensor according to the target pedestrian image, determines target pedestrian area；
Calculating sub module, for calculating at least one object vector feature corresponding with the target pedestrian area；
Acquisition submodule, for based at least one described object vector feature, obtaining the target pedestrian characteristic information.
The device described in second aspect is implemented for the pedestrian detection of aforementioned first aspect and knows method for distinguishing again.
According to the third aspect of the invention we there is provided a kind of computer chip, the computer chip includes processor and deposited
Reservoir.The memory storage has instruction code, and the processor is used to perform the instruction code, and when the processor is held
During row instruction code, the pedestrian detection described in aforementioned first aspect can be realized and method for distinguishing is known again.
As can be seen here, the embodiment of the present invention can provide the characteristic information for ReID while pedestrian detection is carried out.
The efficiency of processing can not only be so improved, and can avoid introducing extra error, it is ensured that the precision of processing.
Brief description of the drawings
By the way that the embodiment of the present invention is described in more detail with reference to accompanying drawing, above-mentioned and other purpose of the invention,
Feature and advantage will be apparent.Accompanying drawing is used for providing further understanding the embodiment of the present invention, and constitutes explanation
A part for book, is used to explain the present invention together with the embodiment of the present invention, is not construed as limiting the invention.In the accompanying drawings,
Identical reference number typically represents same parts or step.
Fig. 1 is a schematic block diagram of the electronic equipment of the embodiment of the present invention；
Fig. 2 is the pedestrian detection of the embodiment of the present invention and knows an indicative flowchart of method for distinguishing again；
Fig. 3 is the pedestrian detection of the embodiment of the present invention and another indicative flowchart for knowing method for distinguishing again；
Fig. 4 is the pedestrian detection of the embodiment of the present invention and a schematic block diagram of the device recognized again.
Embodiment
Become apparent in order that obtaining the object, technical solutions and advantages of the present invention, root is described in detail below with reference to accompanying drawings
According to the example embodiment of the present invention.Obviously, described embodiment is only a part of embodiment of the present invention, rather than this hair
Bright whole embodiments, it should be appreciated that the present invention is not limited by example embodiment described herein.Described in the present invention
The embodiment of the present invention, those skilled in the art's all other embodiment resulting in the case where not paying creative work
It should all fall under the scope of the present invention.
The embodiment of the present invention can apply to electronic equipment, and Fig. 1 show one of the electronic equipment of the embodiment of the present invention
Schematic block diagram.Electronic equipment 10 shown in Fig. 1 include one or more processors 102, one or more storage devices 104,
Input unit 106, output device 108, imaging sensor 110 and one or more non-image sensors 114, these components lead to
Cross bus system 112 and/or other forms interconnection.It should be noted that the component and structure of electronic equipment 10 shown in Fig. 1 are to show
Example property, and it is nonrestrictive, and as needed, the electronic equipment can also have other assemblies and structure.
The processor 102 can include CPU 1021 and GPU 1022 or with data-handling capacity and/or instruction
The processing unit of the other forms of executive capability, such as field programmable gate array (Field-Programmable Gate
Array, FPGA) or advanced reduced instruction set machine (Advanced RISC (Reduced Instruction Set
Computer) Machine, ARM) etc., and processor 102 can control other components in the electronic equipment 10 to perform
Desired function.
The storage device 104 can include one or more computer program products, and the computer program product can
With including various forms of computer-readable recording mediums, such as volatile memory 1041 and/or nonvolatile memory
1042.The volatile memory 1041 can for example include random access memory (Random Access Memory, RAM)
And/or cache memory (cache) etc..The nonvolatile memory 1042 can for example include read-only storage
(Read-Only Memory, ROM), hard disk, flash memory etc..One or many can be stored on the computer-readable recording medium
Individual computer program instructions, processor 102 can run described program instruction, to realize various desired functions.In the meter
Can also store various application programs and various data in calculation machine readable storage medium storing program for executing, such as application program use and/or
Various data produced etc..
The input unit 106 can be that user is used for the device of input instruction, and can include keyboard, mouse, wheat
One or more of gram wind and touch-screen etc..
The output device 108 can export various information (such as image or sound) to outside (such as user), and
One or more of display, loudspeaker etc. can be included.
Described image sensor 110 can shoot the desired image of user (such as photo, video), and will be captured
Image be stored in the storage device 104 so that other components are used.
Work as attention, the component and structure of the electronic equipment 10 shown in Fig. 1 are exemplary, although the electronics shown in Fig. 1
Equipment 10 includes multiple different devices, but as needed, some of which device can not be necessary, some of which
The quantity of device can be of the invention that this is not limited with more etc..
Fig. 2 is an indicative flowchart for knowing method for distinguishing again for the pedestrian detection of the embodiment of the present invention.Shown in Fig. 2
Method includes：
S101, extracts the characteristic tensor of original image.
As one, original image can be the image gathered in real time, such as in being the video gathered by camera
A two field picture or multiple image, or, can be the pictures photographed by camera.It is used as another example, original image
Can be obtained from specific source, for example, the image for previously gathering and storing can be obtained from memory.
In S101, original image can be input to one or more layers convolutional neural networks (Convolutional
Neural Network, CNN), extract the characteristic tensor of the original image.
If it is understood that the original image include multiple image, for example, the original image be video, then, can in S101
So that each frame in multiple image is input into one or more layers CNN, so as to extract the characteristic tensor of each two field picture.
Convolutional neural networks (CNN) are a kind of feedforward neural networks, are made up of some convolution units.Each convolution unit can
To respond the surrounding cells in a part of coverage.The parameter of each convolution unit is optimized by back-propagation algorithm
Arrive.The purpose of convolution algorithm is to extract the different characteristic of input.For example, first layer convolutional layer may can only to extract some rudimentary
Feature, such as edge, lines and angle level, the network of more layers can from low-level features the more complicated feature of iterative extraction.
These features are similar to thermodynamic chart, for an image (frame in such as video), the feature extracted by convolutional neural networks
Tensor can be expressed as a three rank tensor X.Three dimensions of the three ranks tensor represent transverse direction (H), longitudinal direction (W) and channel respectively
(C), wherein, channel (C) is to include original image in the dimension that is set according to the demand of application scenarios, channel (C)
Feature, because the species of the primitive image features required for different usage scenarios is different from fine degree, therefore by different
Convolutional neural networks obtained by channel (C) might not be identical, it is set according to different demands.Alternatively, make
For one, after convolutional neural networks are according to being trained the need for application scenarios, a pictures of input are calculated, are based on
Length information, width information and the color channel information for inputting picture export a three rank tensors, and three dimensions represent horizontal stroke respectively
To (H), longitudinal direction (W) and channel (C), above-mentioned each dimension is comprising 128 numerical value.That is, obtained by S101 extractions
The dimension of characteristic tensor can be expressed as H × W × C.Element in this feature tensor can be expressed as X [i] [j] [k], wherein, i
=0,1,2 ..., H-1, j=0,1,2 ..., W-1, k=0,1,2 ..., C-1.
Exemplarily, the parameter of the convolutional neural networks used in S101 can random initializtion obtain, also may be used
To be that the network trained before carries out initializing what is obtained, wherein the network trained such as AlexNet,
VGG, residual error network (Residential Network, ResNet) etc..
In the embodiment of the present invention, some of these networks trained part can be chosen as being made in S101
A part for convolutional neural networks；Or, a part of parameter in the network that these have trained can also be fixed, and it is right
Other parameters are trained, so as to obtain the parameter of the convolutional neural networks used in S101.
S102, according to the characteristic tensor, determines at least one subregion.
Exemplarily, S102 can include：According to the characteristic tensor, multiple super-pixel points, each super-pixel point are built
Represent a C dimensional vector；According to the multiple super-pixel point, it is determined that with the one-to-one multiple areas of the multiple super-pixel point
Domain；At least one subregion according to being determined the multiple region.
Specifically, X [i] [j] can be defined for super-pixel point, it will be understood that a super-pixel point be a C tie up to
Amount.One super-pixel point can correspond to the panel region on original image.Specifically, X [i] [j] represent be in tensor abscissa
I, the C dimensional vector that ordinate constitutes for the numerical value of all channels on j position.It is understood that in S102, can be according to spy
Levy tensor and define H × W super-pixel point.Wherein, the vector of each super-pixel point contains the corresponding region of super-pixel point
Feature.That is, each super-pixel point represents a region, multiple super-pixel points are corresponded with multiple regions.
Alternatively, as one, at least one subregion according to being determined the multiple region can include：Using
Non-maxima suppression (Non-Maximum Suppression, NMS) algorithm, based at least one described in the determination of the multiple region
Sub-regions.
Specifically, by by the vector of each super-pixel point be input to a grader and return device, can be derived that with it is defeated
It whether there is the object of some particular size, and object rectangle region that may be present in the corresponding region of super-pixel point entered
Domain.All super-pixel points are input to after grader and recurrence device, can obtain a large amount of possible objects and correspondence rectangle region
Domain.Further, to these substantial amounts of rectangular areas, according to the reliability order of grader, and the higher square of degree of overlapping is merged
Shape region, just can finally obtain at least one subregion.Here at least one subregion, which is considered, there may be sense
The region of interest object.Exemplarily, grader and recurrence device can be 1 × 1 convolutional networks.
Alternatively, as another example, at least one subregion according to being determined the multiple region can include：Adopt
With non-maxima suppression (Non-Maximum Suppression, NMS) algorithm, multiple rectangles are determined based on the multiple region
Region；The corresponding feature in the multiple rectangular area is integrated, at least one described subregion is obtained.
Specifically, by by the vector of each super-pixel point be input to a grader and return device, can be derived that with it is defeated
It whether there is the object of some particular size, and object rectangle region that may be present in the corresponding region of super-pixel point entered
Domain.All super-pixel points are input to after grader and recurrence device, can obtain a large amount of possible objects and correspondence rectangle region
Domain, wherein, the region of the rectangular area obtained by grader and after returning device and the super-pixel point of input in artwork is not
Correspond, the rectangular area is typically larger than corresponding region of the super-pixel point in artwork, therefore obtained rectangular area is usual
Corresponding at least one super-pixel point.Then, to these substantial amounts of rectangular areas, according to the reliability order of grader, and close
And the higher rectangular area of degree of overlapping, it just can finally obtain multiple rectangular areas.Here multiple rectangular areas may be considered that
It is the region that there may be attention object.Exemplarily, grader and recurrence device can be 1 × 1 convolutional networks.Enter one
Step ground, integrates multiple rectangular areas (i.e. these regions that there is attention object) corresponding feature, and such as region is drawn
It is divided into some subregions (being for example divided into multiple grids, for example, be divided into N × M grid, such as 3 × 6, or 1 × 1), so that
At least one subregion is obtained, by being some subregions region division, the corresponding vector in region finally given can be made
Feature specification is consistent, is convenient for unified operation.
It is understood that in the embodiment of the present invention, substantial amounts of rectangular area is merged into used strategy for NMS, NMS can
To be merged by two indices：Confidence level and degree of overlapping.Exemplarily, grader can provide a score value (score)
To represent the confidence level of current candidate frame, the higher candidate frame of confidence level can be retained as far as possible when merging.Exemplarily, it is overlapping
Degree is referred to as overlapping area (Intersection-over Union, IoU), if the candidate frame overlapping area being positioned adjacent to
It can then be merged more than certain threshold value.
S103, is calculated and at least one one-to-one vector characteristics of at least one described subregion.
Specifically, for every sub-regions at least one described subregion：Will be all in every sub-regions
The corresponding vector of super-pixel point in each channel take maximum or average value, obtain and described per the corresponding vector of sub-regions
Feature.
Exemplarily, each channel in the vector corresponding to corresponding each super-pixel point in every sub-regions can be taken
Maximum or average value etc. obtain the corresponding vector of every sub-regions, will merge or connect into one by corresponding vector per sub-regions
Individual long vector is as obtaining the corresponding vector characteristics in the region.It is understood that the channel of vector is its corresponding super-pixel point
Corresponding c dimensional vectors.
S104, based at least one described vector characteristics, determines that the position of pedestrian and extraction are used in the original image
In the pedestrian's characteristic information to be identified recognized again.
S104 can include：Based at least one described vector characteristics, the position of pedestrian in the original image is determined, its
In, the positional representation is coordinate of the pedestrian in the original image；And based at least one described vector characteristics,
Pedestrian's characteristic information to be identified for recognizing again is extracted, wherein, pedestrian's characteristic information to be identified is in original image
Pedestrian's characteristic information of each pedestrian extracted, that is to say, that the row of each pedestrian extracted in original image
People's characteristic information may be defined as pedestrian's characteristic information to be identified.Multiple rows are generally comprised in actual scene, in original image
People, by the present embodiment, can extract multiple pedestrian's characteristic informations, and multiple pedestrian's features to extracting in original image
Information is handled.
So, the characteristic information required for ReID being obtained while pedestrian detection is carried out, it is possible to increase processing
Efficiency, introduces extra error while being avoided that, improves precision.
Exemplarily, in S104, it can also include：Based at least one described vector characteristics, judge described at least one
Whether the object in sub-regions is pedestrian.
Specifically, vector characteristics can be separately input in three different graders/recurrence devices.First be used for pair
Object is detected whether judge the object is pedestrian.Second is used to obtain the particular location where the object, that is, obtains one
The individual rectangle frame for closely surrounding the object.3rd is used for the characteristic information to the Object Extraction available for ReID.
Exemplarily, as shown in figure 3, vector characteristics 30 can be based on, concurrently determine whether object is pedestrian 32, object
Position 34 and object ReID characteristic information 36.
Wherein it is possible to using object classify (Object Classifier) object in subregion is detected, to sentence
Whether the disconnected object is pedestrian.
Wherein it is possible to return the position that (Bounding Box Regressor) obtains the object in subregion using frame
Put, such as can be the coordinate for the rectangle frame for closely surrounding the object.For example, the position can include the upper left corner and the right side of rectangle frame
The coordinate value of inferior horn, or, include the lower left corner and the coordinate value in the upper right corner of rectangle frame, or, include the lower left corner of rectangle frame
Coordinate value and rectangle frame length and width value.
Used wherein it is possible to return (ReID Feature Regressor) using ReID features and obtain the object in subregion
In ReID characteristic information.
Exemplarily, if determining that the object is not pedestrian 32, then can be by corresponding position 34 and corresponding
ReID characteristic information 36 is rejected.
Exemplarily, if determining that the object is pedestrian 32, then can be by corresponding position 34 and corresponding ReID
Characteristic information 36 retain.At this point it is possible to understand, resulting position 34 is the position of pedestrian, and resulting is used for ReID's
Characteristic information 36 is the characteristic information that is extracted to pedestrian.
Alternatively, as one embodiment, after the method shown in Fig. 2, it can also include：It is used to recognize again by described
Pedestrian's characteristic information to be identified be compared with target pedestrian's characteristic information, to judge the row to be identified for being used to recognize again
Whether the corresponding pedestrian of people's characteristic information and the target pedestrian are same people.Target pedestrian's characteristic information is obtained by following steps
：Extract the characteristic tensor of target pedestrian image；Target pedestrian area is determined according to the characteristic tensor of the target pedestrian image；
At least one object vector feature corresponding with the target pedestrian area is calculated, it is special based at least one described object vector
Levy, obtain the target pedestrian characteristic information.
Wherein, target pedestrian characteristic information can be directed to target image, obtained by similar to the method shown in Fig. 2
's.Target pedestrian's characteristic information is referred to as known features information.Exemplarily, target pedestrian image can be primarily based on true
The rectangular area that the target pedestrian being scheduled in above-mentioned image is present, calculates vector characteristics corresponding with the rectangular area, and be based on
The vector characteristics determine the ReID of target pedestrian characteristic information, are used as target pedestrian's characteristic information.Specifically, the process institute
The neutral net used can be with identical with the neutral net employed in the embodiment shown in earlier figures 2.
Exemplarily, if for the point between the pedestrian's characteristic information to be identified and target pedestrian's characteristic information that recognize again
The result multiplied is more than or equal to default threshold value, then can determine represented by the pedestrian's characteristic information to be identified for being used to recognize again
The artificial same people of pedestrian and target line.
For example, x will can be expressed as the pedestrian's characteristic information to be identified recognized againp, by target pedestrian's feature
Information is expressed as xqIf, xp·xqMore than or equal to default threshold value, then illustrate xpRepresented pedestrian is artificially same with target line
People.Or, if | xp-xq| more than or equal to default another threshold value, then illustrate xpRepresented pedestrian is artificially same with target line
People.Correspondingly, it is appreciated that if xp·xqLess than default threshold value, or, if | xp-xq| less than default another threshold value, then
Illustrate xpRepresented pedestrian and target pedestrian are not same people.
Exemplarily, can find in the characteristic information in S104 with maximum that of the similarity of target pedestrian's characteristic information
Individual characteristic information, and determine the pedestrian represented by the characteristic information found and the artificial same people of target line.
For example, the dot product between all characteristic informations and target pedestrian's characteristic information for ReID is calculated, such as
That characteristic information for being used for ReID for the result maximum that fruit dot multiplies is xp, it is determined that xpRepresented pedestrian and target line are artificial
Same people.Or, the difference between all characteristic informations and target pedestrian's characteristic information for ReID is calculated, should calculating
The mould of difference, if that maximum characteristic information for being used for ReID of the result of mould is xp, it is determined that xpRepresented pedestrian and mesh
Mark pedestrian is same people.
In addition, in the embodiment of the present invention, in S104, can using the good neutral net of training in advance, based on it is described extremely
Few vector characteristics, extract the characteristic information for ReID.
Wherein, the parameter of the good neutral net of training in advance can be obtained using following method training：
(a) vector corresponding to the image of two people is provided：xiAnd xj, obtain so that function ∑ (1-1/2 Δs (xi,xj))
(xi·xj) take minimum value neutral net parameter.
Wherein, if the two artificial same persons, Δ (xi,xj)=1；If the two artificial different people, Δ
(xi,xj)=0.
(b) vector corresponding to the image of three people is provided：xi、xjAnd xk, obtain so that function ∑ (xi·xk-xi·xj)
Take the parameter of the neutral net of minimum value.
Wherein, xiAnd xjIt is characteristic vector of the same person under different scenes, xkFor the characteristic vector of another person.xi
And xkFor under same scene, or xjAnd xkFor under same scene.
It can be seen that, in the embodiment of the present invention, it can be declined by back-propagation and gradient, train whole end2end
Detect network and ReID networks.
As can be seen here, the embodiment of the present invention can use convolutional neural networks (CNN), in inspection for an input picture
While measuring all pedestrians in image, giving these people is used for ReID characteristic information.Processing can not only so be improved
Efficiency, and can avoid introducing extra error, it is ensured that the precision of processing.
Fig. 4 is the pedestrian detection of the embodiment of the present invention and a schematic block diagram of the device recognized again.Dress shown in Fig. 4
Putting 40 includes：Extraction module 401, the first determining module 402, the determining module 404 of computing module 403 and second.
Extraction module 401, the characteristic tensor for extracting original image；
First determining module 402, for the characteristic tensor extracted according to extraction module 401, determines at least one son
Region；
Computing module 403, is corresponded for calculating with least one subregion described in the determination of the first determining module 402
At least one vector characteristics；
Second determining module 404, at least one vector characteristics described in being obtained based on computing module 403, determines institute
State the position of pedestrian in original image and extract pedestrian's characteristic information to be identified for recognizing again.
Exemplarily, the first determining module 402 can include：Submodule, the first determination sub-module and second is built to determine
Submodule.
Submodule is built, for according to the characteristic tensor, building multiple super-pixel points, each super-pixel point represents one
C dimensional vectors；
First determination sub-module, for according to the multiple super-pixel point, it is determined that with a pair of the multiple super-pixel point 1
The multiple regions answered；
Second determination sub-module, at least one subregion according to the determination of the multiple region.
Exemplarily, second determination sub-module can be used for：Using non-maxima suppression NMS algorithms, based on described
Multiple regions determine multiple rectangular areas；The corresponding feature in the multiple rectangular area is integrated, described at least one is obtained
Sub-regions.
Exemplarily, computing module 403 can be used for：For every sub-regions at least one described subregion：
Each channel in the corresponding vector of all super-pixel points in every sub-regions is taken into maximum or average
Value, is obtained and the vector characteristics corresponding per sub-regions.
Exemplarily, the second determining module 404 can be used for：
Based at least one described vector characteristics, the position of pedestrian in the original image is determined, wherein, the position table
It is shown as coordinate of the pedestrian in the original image；And
Based at least one described vector characteristics, pedestrian's characteristic information to be identified for recognizing again is extracted.
Exemplarily, the second determining module 404 can be also used for：Based at least one described vector characteristics, judge described
Whether the object at least one subregion is pedestrian.
Exemplarily, judge module can also be included, be used for：By pedestrian's characteristic information to be identified for recognizing again
It is compared with target pedestrian's characteristic information, to judge the corresponding pedestrian of pedestrian's characteristic information to be identified for being used to recognize again
Whether it is same people with the target pedestrian.
Exemplarily, acquisition module can also be included, for the target pedestrian characteristic information to be obtained ahead of time.
Exemplarily, acquisition module can include：
Extracting sub-module, the characteristic tensor for extracting target pedestrian image；
Determination sub-module, for the characteristic tensor according to the target pedestrian image, determines target pedestrian area；
Calculating sub module, for calculating at least one object vector feature corresponding with the target pedestrian area；
Acquisition submodule, for based at least one described object vector feature, obtaining the target pedestrian characteristic information.
Device 40 shown in Fig. 4 is implemented for earlier figures 2 or pedestrian detection shown in Fig. 3 and knows method for distinguishing again.
In addition, the embodiment of the present invention additionally provides another pedestrian detection and the device recognized again, the device can include
Processor, memory and it is stored in the computer program run on the memory and on the processor, computing device
The step of method shown in earlier figures 2 or Fig. 3 is realized during described program.
In addition, the embodiment of the present invention additionally provides a kind of electronic equipment, the electronic equipment can include the device shown in Fig. 4
40.The electronic equipment can realize earlier figures 2 or pedestrian detection shown in Fig. 3 and know method for distinguishing again.
In addition, the embodiment of the present invention additionally provides a kind of computer-readable storage medium, computer program is stored thereon with.Work as institute
State computer program by computing device when, it is possible to achieve shown in earlier figures 2 or Fig. 3 the step of method.For example, the computer is deposited
Storage media is computer-readable recording medium.
As can be seen here, the embodiment of the present invention can provide the characteristic information for ReID while pedestrian detection is carried out.
The efficiency of processing can not only be so improved, and can avoid introducing extra error, it is ensured that the precision of processing.
Although describing example embodiment by reference to accompanying drawing here, it should be understood that above-mentioned example embodiment is merely exemplary
, and be not intended to limit the scope of the invention to this.Those of ordinary skill in the art can carry out various changes wherein
And modification, it is made without departing from the scope of the present invention and spirit.All such changes and modifications are intended to be included in appended claims
Within required the scope of the present invention.
Those of ordinary skill in the art are it is to be appreciated that the list of each example described with reference to the embodiments described herein
Member and algorithm steps, can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
Performed with hardware or software mode, depending on the application-specific and design constraint of technical scheme.Professional and technical personnel
Described function can be realized using distinct methods to each specific application, but this realization is it is not considered that exceed
The scope of the present invention.
, can be by it in several embodiments provided herein, it should be understood that disclosed apparatus and method
Its mode is realized.For example, apparatus embodiments described above are only schematical, for example, the division of the unit, only
Only a kind of division of logic function, can there is other dividing mode when actually realizing, such as multiple units or component can be tied
Another equipment is closed or is desirably integrated into, or some features can be ignored, or do not perform.
In the specification that this place is provided, numerous specific details are set forth.It is to be appreciated, however, that the implementation of the present invention
Example can be put into practice in the case of these no details.In some instances, known method, structure is not been shown in detail
And technology, so as not to obscure the understanding of this description.
Similarly, it will be appreciated that in order to simplify the present invention and help to understand one or more of each inventive aspect, exist
To the present invention exemplary embodiment description in, each feature of the invention be grouped together into sometimes single embodiment, figure,
Or in descriptions thereof.However, the method for the invention should be construed to reflect following intention：It is i.e. claimed
Application claims features more more than the feature being expressly recited in each claim.More precisely, such as corresponding power
As sharp claim reflects, its inventive point is that the spy of all features less than some disclosed single embodiment can be used
Levy to solve corresponding technical problem.Therefore, it then follows it is specific that thus claims of embodiment are expressly incorporated in this
Embodiment, wherein each claim is in itself as the separate embodiments of the present invention.
It will be understood to those skilled in the art that in addition to mutually exclusive between feature, any combinations pair can be used
All features and so disclosed any method disclosed in this specification (including adjoint claim, summary and accompanying drawing)
Or all processes or unit of equipment are combined.Unless expressly stated otherwise, this specification (including adjoint right will
Ask, make a summary and accompanying drawing) disclosed in each feature can be by offer is identical, equivalent or the alternative features of similar purpose are replaced.
Although in addition, it will be appreciated by those of skill in the art that some embodiments described herein include other embodiments
In included some features rather than further feature, but the combination of the feature of be the same as Example does not mean in of the invention
Within the scope of and form different embodiments.For example, in detail in the claims, embodiment claimed it is one of any
Mode it can use in any combination.
The present invention all parts embodiment can be realized with hardware, or with one or more processor run
Software module realize, or realized with combinations thereof.It will be understood by those of skill in the art that can use in practice
Microprocessor or digital signal processor (DSP) realize some moulds in article analytical equipment according to embodiments of the present invention
The some or all functions of block.The present invention is also implemented as the part or complete for performing method as described herein
The program of device (for example, computer program and computer program product) in portion.Such program for realizing the present invention can be stored
On a computer-readable medium, or can have one or more signal form.Such signal can be from internet
Download and obtain on website, either provide or provided in any other form on carrier signal.
It should be noted that the present invention will be described rather than limits the invention for above-described embodiment, and ability
Field technique personnel can design alternative embodiment without departing from the scope of the appended claims.In the claims,
Any reference symbol between bracket should not be configured to limitations on claims.Word "comprising" is not excluded the presence of not
Element or step listed in the claims.Word "a" or "an" before element does not exclude the presence of multiple such
Element.The present invention can be by means of including the hardware of some different elements and coming real by means of properly programmed computer
It is existing.In if the unit claim of equipment for drying is listed, several in these devices can be by same hardware branch
To embody.The use of word first, second, and third does not indicate that any order.These words can be explained and run after fame
Claim.
The foregoing is only a specific embodiment of the invention or to the explanation of embodiment, protection of the invention
Scope is not limited thereto, any one skilled in the art the invention discloses technical scope in, can be easily
Expect change or replacement, should all be included within the scope of the present invention.Protection scope of the present invention should be with claim
Protection domain is defined.
Claims (16)
1. a kind of pedestrian detection and method for distinguishing is known again, it is characterised in that including：
Extract the characteristic tensor of original image；
According to the characteristic tensor, at least one subregion is determined；
Calculate and at least one one-to-one vector characteristics of at least one described subregion；
Based at least one described vector characteristics, determine the position of pedestrian in the original image and extract for recognizing again
Pedestrian's characteristic information to be identified.
2. the method as described in claim 1, it is characterised in that described according to the characteristic tensor, determines at least one sub-district
Domain, including：
According to the characteristic tensor, multiple super-pixel points are built, each super-pixel point represents a C dimensional vector；
According to the multiple super-pixel point, it is determined that with the one-to-one multiple regions of the multiple super-pixel point；
At least one subregion according to being determined the multiple region.
3. method as claimed in claim 2, it is characterised in that described at least one son according to being determined the multiple region
Region, including：
Using non-maxima suppression NMS algorithms, multiple rectangular areas are determined based on the multiple region；
The corresponding feature in the multiple rectangular area is integrated, at least one described subregion is obtained.
4. method as claimed in claim 3, it is characterised in that the calculating and at least one described subregion are one-to-one
At least one vector characteristics, including：
For every sub-regions at least one described subregion：
Each channel in the corresponding vector of all super-pixel points in every sub-regions is taken into maximum or average value, obtained
To vector characteristics corresponding with the often sub-regions.
5. the method as described in claim 1, it is characterised in that described based at least one described vector characteristics, it is determined that described
The position of pedestrian and pedestrian's characteristic information to be identified for recognizing again is extracted in original image, including：
Based at least one described vector characteristics, the position of pedestrian in the original image is determined, wherein, the positional representation is
Coordinate of the pedestrian in the original image；And
Based at least one described vector characteristics, pedestrian's characteristic information to be identified for recognizing again is extracted.
6. the method as described in claim 1, it is characterised in that also include：
Based at least one described vector characteristics, judge whether the object at least one described subregion is pedestrian.
7. the method as any one of claim 1 to 6, it is characterised in that also include：
Pedestrian's characteristic information to be identified for recognizing again is compared with target pedestrian's characteristic information, it is described to judge
Whether the corresponding pedestrian of pedestrian's characteristic information to be identified and the target pedestrian for recognizing again are same people.
8. method as claimed in claim 7, it is characterised in that the target pedestrian characteristic information is obtained by following steps：
Extract the characteristic tensor of target pedestrian image；
According to the characteristic tensor of the target pedestrian image, target pedestrian area is determined；
At least one object vector feature corresponding with the target pedestrian area is calculated, based at least one described object vector
Feature, obtains the target pedestrian characteristic information.
9. a kind of pedestrian detection and the device recognized again, it is characterised in that including：
Extraction module, the characteristic tensor for extracting original image；
First determining module, for according to the characteristic tensor, determining at least one subregion；
Computing module, for calculating and at least one one-to-one vector characteristics of at least one described subregion；
Second determining module, for based at least one described vector characteristics, determine the position of pedestrian in the original image with
And extract pedestrian's characteristic information to be identified for recognizing again.
10. device as claimed in claim 9, it is characterised in that first determining module, including：
Submodule is built, for according to the characteristic tensor, building multiple super-pixel points, each super-pixel point represents a C dimension
Vector；
First determination sub-module, for according to the multiple super-pixel point, it is determined that one-to-one with the multiple super-pixel point
Multiple regions；
Second determination sub-module, at least one subregion according to the determination of the multiple region.
11. device as claimed in claim 10, it is characterised in that second determination sub-module, is used for：
Using non-maxima suppression NMS algorithms, multiple rectangular areas are determined based on the multiple region；
The corresponding feature in the multiple rectangular area is integrated, at least one described subregion is obtained.
12. device as claimed in claim 11, it is characterised in that the computing module, is used for：
For every sub-regions at least one described subregion：
Each channel in the corresponding vector of all super-pixel points in every sub-regions is taken into maximum or average value, obtained
To vector characteristics corresponding with the often sub-regions.
13. device as claimed in claim 9, it is characterised in that second determining module, is used for：
Based at least one described vector characteristics, the position of pedestrian in the original image is determined, wherein, the positional representation is
Coordinate of the pedestrian in the original image；And
Based at least one described vector characteristics, pedestrian's characteristic information to be identified for recognizing again is extracted.
14. device as claimed in claim 9, it is characterised in that second determining module, is additionally operable to：
Based at least one described vector characteristics, judge whether the object at least one described subregion is pedestrian.
15. the device as any one of claim 9 to 14, it is characterised in that also including judge module, is used for：
Pedestrian's characteristic information to be identified for recognizing again is compared with target pedestrian's characteristic information, it is described to judge
Whether the corresponding pedestrian of pedestrian's characteristic information to be identified and the target pedestrian for recognizing again are same people.
16. device as claimed in claim 15, it is characterised in that also including acquisition module, for the target to be obtained ahead of time
Pedestrian's characteristic information；
Wherein, the acquisition module includes：
Extracting sub-module, the characteristic tensor for extracting target pedestrian image；
Determination sub-module, for the characteristic tensor according to the target pedestrian image, determines target pedestrian area；
Calculating sub module, for calculating at least one object vector feature corresponding with the target pedestrian area；
Acquisition submodule, for based at least one described object vector feature, obtaining the target pedestrian characteristic information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710330307.1A CN106971178A (en) | 2017-05-11 | 2017-05-11 | Pedestrian detection and the method and device recognized again |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710330307.1A CN106971178A (en) | 2017-05-11 | 2017-05-11 | Pedestrian detection and the method and device recognized again |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106971178A true CN106971178A (en) | 2017-07-21 |
Family
ID=59330691
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710330307.1A Pending CN106971178A (en) | 2017-05-11 | 2017-05-11 | Pedestrian detection and the method and device recognized again |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106971178A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108062562A (en) * | 2017-12-12 | 2018-05-22 | 北京图森未来科技有限公司 | A kind of object recognition methods and device again |
| CN108875507A (en) * | 2017-11-22 | 2018-11-23 | 北京旷视科技有限公司 | Pedestrian tracting method, equipment, system and computer readable storage medium |
| CN108875500A (en) * | 2017-11-06 | 2018-11-23 | 北京旷视科技有限公司 | Pedestrian recognition methods, device, system and storage medium again |
| CN108875505A (en) * | 2017-11-14 | 2018-11-23 | 北京旷视科技有限公司 | Pedestrian neural network based recognition methods and device again |
| CN109670555A (en) * | 2018-12-27 | 2019-04-23 | 吉林大学 | Instance-level pedestrian detection and pedestrian's weight identifying system based on deep learning |
| 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 |
| CN109978043A (en) * | 2019-03-19 | 2019-07-05 | 新华三技术有限公司 | A kind of object detection method and device |
| CN109977928A (en) * | 2019-04-25 | 2019-07-05 | 中国科学院自动化研究所 | A kind of robot target pedestrian retrieval method |
| US10552979B2 (en) | 2017-09-13 | 2020-02-04 | TuSimple | Output of a neural network method for deep odometry assisted by static scene optical flow |
| CN111091091A (en) * | 2019-12-16 | 2020-05-01 | 北京迈格威科技有限公司 | Method, device and equipment for extracting target object re-identification features and storage medium |
| US10671083B2 (en) | 2017-09-13 | 2020-06-02 | Tusimple, Inc. | Neural network architecture system for deep odometry assisted by static scene optical flow |
| WO2020151329A1 (en) * | 2019-01-23 | 2020-07-30 | 平安科技（深圳）有限公司 | Target detection based identification box determining method and device and terminal equipment |
| CN111597847A (en) * | 2019-02-20 | 2020-08-28 | 中科院微电子研究所昆山分所 | Two-dimensional code identification method, device and equipment and readable storage medium |
| US10762635B2 (en) | 2017-06-14 | 2020-09-01 | Tusimple, Inc. | System and method for actively selecting and labeling images for semantic segmentation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7123745B1 (en) * | 1999-11-24 | 2006-10-17 | Koninklijke Philips Electronics N.V. | Method and apparatus for detecting moving objects in video conferencing and other applications |
| CN105160319A (en) * | 2015-08-31 | 2015-12-16 | 电子科技大学 | Method for realizing pedestrian re-identification in monitor video |
| CN105184297A (en) * | 2015-08-24 | 2015-12-23 | 西安电子科技大学 | Polarized SAR image classification method based on tensor and sparse self-coder |
| CN106529511A (en) * | 2016-12-13 | 2017-03-22 | 北京旷视科技有限公司 | Image structuring method and device |
-
2017
- 2017-05-11 CN CN201710330307.1A patent/CN106971178A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7123745B1 (en) * | 1999-11-24 | 2006-10-17 | Koninklijke Philips Electronics N.V. | Method and apparatus for detecting moving objects in video conferencing and other applications |
| CN105184297A (en) * | 2015-08-24 | 2015-12-23 | 西安电子科技大学 | Polarized SAR image classification method based on tensor and sparse self-coder |
| CN105160319A (en) * | 2015-08-31 | 2015-12-16 | 电子科技大学 | Method for realizing pedestrian re-identification in monitor video |
| CN106529511A (en) * | 2016-12-13 | 2017-03-22 | 北京旷视科技有限公司 | Image structuring method and device |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10762635B2 (en) | 2017-06-14 | 2020-09-01 | Tusimple, Inc. | System and method for actively selecting and labeling images for semantic segmentation |
| US10552979B2 (en) | 2017-09-13 | 2020-02-04 | TuSimple | Output of a neural network method for deep odometry assisted by static scene optical flow |
| US10671083B2 (en) | 2017-09-13 | 2020-06-02 | Tusimple, Inc. | Neural network architecture system for deep odometry assisted by static scene optical flow |
| CN108875500A (en) * | 2017-11-06 | 2018-11-23 | 北京旷视科技有限公司 | Pedestrian recognition methods, device, system and storage medium again |
| CN108875500B (en) * | 2017-11-06 | 2022-01-07 | 北京旷视科技有限公司 | Pedestrian re-identification method, device and system and storage medium |
| CN108875505A (en) * | 2017-11-14 | 2018-11-23 | 北京旷视科技有限公司 | Pedestrian neural network based recognition methods and device again |
| CN108875507A (en) * | 2017-11-22 | 2018-11-23 | 北京旷视科技有限公司 | Pedestrian tracting method, equipment, system and computer readable storage medium |
| CN108875507B (en) * | 2017-11-22 | 2021-07-23 | 北京旷视科技有限公司 | Pedestrian tracking method, apparatus, system, and computer-readable storage medium |
| CN108062562A (en) * | 2017-12-12 | 2018-05-22 | 北京图森未来科技有限公司 | A kind of object recognition methods and device again |
| US11093789B2 (en) | 2017-12-12 | 2021-08-17 | Tusimple, Inc. | Method and apparatus for object re-identification |
| CN108062562B (en) * | 2017-12-12 | 2020-03-10 | 北京图森未来科技有限公司 | Object re-recognition method and device |
| 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 |
| CN109670555A (en) * | 2018-12-27 | 2019-04-23 | 吉林大学 | Instance-level pedestrian detection and pedestrian's weight identifying system based on deep learning |
| CN109670555B (en) * | 2018-12-27 | 2023-07-07 | 吉林大学 | Instance-level pedestrian detection and pedestrian re-recognition system based on deep learning |
| WO2020151329A1 (en) * | 2019-01-23 | 2020-07-30 | 平安科技（深圳）有限公司 | Target detection based identification box determining method and device and terminal equipment |
| CN111597847A (en) * | 2019-02-20 | 2020-08-28 | 中科院微电子研究所昆山分所 | Two-dimensional code identification method, device and equipment and readable storage medium |
| CN109978043A (en) * | 2019-03-19 | 2019-07-05 | 新华三技术有限公司 | A kind of object detection method and device |
| CN109977928A (en) * | 2019-04-25 | 2019-07-05 | 中国科学院自动化研究所 | A kind of robot target pedestrian retrieval method |
| CN111091091A (en) * | 2019-12-16 | 2020-05-01 | 北京迈格威科技有限公司 | Method, device and equipment for extracting target object re-identification features and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106971178A (en) | Pedestrian detection and the method and device recognized again | |
| CN111738231B (en) | Target object detection method and device, computer equipment and storage medium | |
| JP7236545B2 (en) | Video target tracking method and apparatus, computer apparatus, program | |
| CN113591872B (en) | A data processing system, object detection method and device | |
| CN109815843B (en) | Image processing method and related product | |
| CN112381061B (en) | Facial expression recognition method and system | |
| CN111259758B (en) | Two-stage remote sensing image target detection method for dense area | |
| CN111738110A (en) | Vehicle target detection method in remote sensing images based on multi-scale attention mechanism | |
| CN109977978B (en) | Multi-target detection method, device and storage medium | |
| CN109492638A (en) | Method for text detection, device and electronic equipment | |
| CN109376681A (en) | A multi-person pose estimation method and system | |
| CN110807362A (en) | Image detection method and device and computer readable storage medium | |
| CN107944403B (en) | Method and device for detecting pedestrian attribute in image | |
| KR102476022B1 (en) | Face detection method and apparatus thereof | |
| CN111524145A (en) | Intelligent picture clipping method and system, computer equipment and storage medium | |
| CN108182695B (en) | Target tracking model training method and device, electronic device and storage medium | |
| CN108446694A (en) | A kind of object detection method and device | |
| CN114119992B (en) | Multi-mode three-dimensional target detection method and device based on fusion of image and point cloud | |
| CN112348116B (en) | Target detection method and device using space context and computer equipment | |
| CN111814754A (en) | Single-frame image pedestrian detection method and device for night scene | |
| CN115496971A (en) | An infrared target detection method, device, electronic equipment and storage medium | |
| CN118608510A (en) | Printed circuit board defect recognition method and system based on BiDC-YOLO model | |
| CN113610798A (en) | Image detection method, image detection device, electronic equipment and storage medium | |
| CN119152185A (en) | Target detection method, device, vehicle and storage medium | |
| CN118887200B (en) | Defect identification method, defect identification device, computer device and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170721 |