CN110427814A - A kind of bicyclist recognition methods, device and equipment again - Google Patents
A kind of bicyclist recognition methods, device and equipment again Download PDFInfo
- Publication number
- CN110427814A CN110427814A CN201910550548.6A CN201910550548A CN110427814A CN 110427814 A CN110427814 A CN 110427814A CN 201910550548 A CN201910550548 A CN 201910550548A CN 110427814 A CN110427814 A CN 110427814A
- Authority
- CN
- China
- Prior art keywords
- people
- vehicle
- vehicle picture
- bicyclist
- feature vector
- 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; 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- 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
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- General Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention provides a kind of bicyclist again recognition methods, device and equipment.By carrying out transfer learning to pedestrian's weight identification model, obtain bicyclist's weight identification model, the feature vector of multiple people's vehicle pictures in people's vehicle picture library is extracted using bicyclist weight identification model, calculate the similarity value between feature vector, it and by the corresponding at least two people's vehicle picture recognition of the highest similarity value is same bicyclist, the matching degree that bicyclist identifies again is improved, realizes the monitoring to non power driven vehicle illegal activities.
Description
Technical field
The present invention relates to field of image processing more particularly to bicyclist's weight identification technologies.
Background technique
With the raising of urbanization rate, the resident population in each cities and towns in the whole nation is being gradually increased, this gives script congested traffic
Bring very big pressure.All kinds of traffic violations behaviors also give related improvement department to make troubles.Wherein non-motor vehicle accounts for motor vehicle
The unlawful practice in road has the characteristics of multiple, normal hair.At present motor vehicle violation identification capture evidence obtaining accomplished substantially high accuracy,
It is automatical and efficient, and the improvement of non-motor vehicle is more inefficient at present.Main reason is that the relevant detection of people vehicle, recognizer are ground
Study carefully very poor.
In actually administering, can set up an office installation monitoring above car lane.By (ride every frame people's vehicle to monitor video
The assembly of passerby and non-motor vehicle) detection, the unlawful practice that all non-motor vehicles account for car lane will be by candid photograph record.
But only accomplish this step under actual conditions not enough, improvement personnel, which need to obtain, opens candid photograph figure with the back side two before violator's vehicle
Piece.Have the facial information of bicyclist before people's vehicle, behind have the license plate information of electric vehicle.After these information can be assisted preferably
Continuous violation disposition.To reach this requirement, monitoring can be carried out mounted in pairs by relevant departments.One camera is towards wagon flow direction
It shoots (back side can be captured), another then shoots (can capture front) in the opposite direction.Either front or the back side, people's car test
All there is no problem for survey.Reluctant problem is that the positive dough figurine vehicle of the lower candid photograph of a monitoring and lower capture of another monitoring are carried on the back
Dough figurine vehicle is matched, i.e., bicyclist identifies again.Under actual scene, manned, object that keeps out the wind etc. can all make the front and back two of people's vehicle
Face difference is huge, and it is difficult that this significantly increases matching.Moreover, can be used for training the data of identification model few.
The matching degree identified again therefore, it is necessary to improve bicyclist.
Summary of the invention
The present invention provides a kind of bicyclist recognition methods, device and equipment again, to improve the matching degree that bicyclist identifies again.
In a first aspect, providing a kind of bicyclist's recognition methods again, comprising:
Transfer learning is carried out to pedestrian's weight identification model, obtains bicyclist's weight identification model；
The feature vector of multiple people's vehicle pictures in people's vehicle picture library is extracted using bicyclist weight identification model；
Calculate multiple similarity values between described eigenvector；
It is same bicyclist by the corresponding at least two people's vehicle picture recognition of the highest similarity value.
In one implementation, described that transfer learning is carried out to pedestrian's weight identification model, obtain bicyclist's weight identification model, packet
It includes:
Obtain the core network parameter of pedestrian's weight identification model；
According to the classification number of bicyclist's training sample, classification is added after the core network of pedestrian weight identification model
Layer；
The parameter and the core network parameter for adjusting the classification layer obtain bicyclist's weight identification model.
In another realization, people's vehicle picture includes the positive dough figurine vehicle figure and people from back side Che Tu of each bicyclist, institute
State the feature vector that multiple people's vehicle pictures in people's vehicle picture library are extracted using bicyclist weight identification model, comprising:
The feature vector of the positive dough figurine vehicle figure and the feature vector of people from back side Che Tu are extracted respectively；
Multiple similarity values between the calculating described eigenvector, comprising:
The feature vector of the feature vector of the positive dough figurine vehicle picture and people from back side vehicle picture is subjected to inner product respectively
Operation obtains the inner product result between the multiple feature vector, wherein the inner product result is as each group of positive dough figurine vehicle figure
Similarity value between the feature vector of piece and the feature vector of people from back side vehicle picture.
In another realization, the method also includes:
The image information of multiple people's vehicle pictures is obtained by monitoring device, wherein the monitoring device includes at least one
The video camera and a video camera to the progress rear shooting of people's vehicle of front shooting, described image packet are carried out to people's vehicle
It includes the candid photograph place of people's vehicle picture and captures the time；
According to the candid photograph place and/or the candid photograph time, people's vehicle picture is stored to people's vehicle of respective type
Picture library；
The feature vector that multiple people's vehicle pictures in people's vehicle picture library are extracted using bicyclist weight identification model,
Include:
Obtain target person vehicle picture library, wherein the target person vehicle picture library is the people Che Tu of user's specified type
Valut；
The spy of multiple people's vehicle pictures in the target person vehicle picture library is extracted using bicyclist weight identification model
Levy vector.
In another realization, the method also includes:
According to the position between the video camera of shooting in front of the progress and the video camera for carrying out rear shooting away from
From the storage time of people's vehicle picture in people's vehicle picture library is arranged；
It, will when the storage time of people's vehicle picture in people's vehicle picture library that the storage time of people's vehicle picture is more than the setting
Storage time is more than people's vehicle picture outbound of the storage time of the setting.
Second aspect provides a kind of bicyclist's weight identification device, comprising:
Study module obtains bicyclist's weight identification model for carrying out transfer learning to pedestrian's weight identification model；
Extraction module, for extracting the spy of multiple people's vehicle pictures in people's vehicle picture library using bicyclist weight identification model
Levy vector；
Computing module, for calculating multiple similarity values between described eigenvector；
Identification module, for being same by the corresponding at least two people's vehicle picture recognition of the highest similarity value
Bicyclist.
In one implementation, people's vehicle picture includes the positive dough figurine vehicle figure and people from back side Che Tu of each bicyclist；
The extraction module is specifically used for extracting the feature vector of the positive dough figurine vehicle figure and people from back side Che Tu respectively
Feature vector；
The computing module is specifically used for the feature vector of the positive dough figurine vehicle picture and the back side people vehicle picture
Feature vector carries out inner product operation respectively, obtains the inner product result between the multiple feature vector, wherein the inner product result
As the similarity value between the feature vector of each group of positive dough figurine vehicle picture and the feature vector of back side people's vehicle picture.
In another realization, described device further include:
Module is obtained, for obtaining the image information of multiple people's vehicle pictures by monitoring device, wherein the monitoring device
The video camera and a video camera to the progress rear shooting of people's vehicle of front shooting, institute are carried out to people's vehicle including at least one
State the candid photograph place and capture the time that image information includes people's vehicle picture；
Memory module, for according to the candid photograph place and/or the candid photograph time, people's vehicle picture to be stored to phase
Answer people's vehicle picture library of type；
The extraction module includes:
First acquisition unit, for obtaining target person vehicle picture library, wherein the target person vehicle picture library is the user
People's vehicle picture library of specified type；
Extraction unit, it is multiple described in the target person vehicle picture library for being extracted using bicyclist weight identification model
The feature vector of people's vehicle picture.
In another realization, the study module includes:
Second acquisition unit, for obtaining the core network parameter of pedestrian's weight identification model；
Adding unit, for the classification number according to bicyclist's training sample, in the backbone network of pedestrian weight identification model
Addition classification layer after network；
Adjustment unit obtains bicyclist's weight for adjusting the parameter and the core network parameter of the classification layer
Identification model.
In another realization, described device further include:
Setup module, for the video camera and the video camera for carrying out rear shooting according to shooting in front of the progress
Between positional distance, the storage time of people's vehicle picture in people's vehicle picture library is set；
The memory module is also used to when people's vehicle in people's vehicle picture library of the storage time of people's vehicle picture more than the setting
It is more than people's vehicle picture outbound of the storage time of the setting by storage time when the storage time of picture.
The third aspect provides a kind of bicyclist and identifies equipment again, including processor, input equipment, output equipment and deposits
Reservoir, the memory is for storing computer program, and the computer program includes program instruction, and the processor is configured
For calling described program to instruct, executes any one of above-mentioned first aspect or first aspect and realize the method.
Fourth aspect provides a kind of computer readable storage medium, is stored in the computer readable storage medium
Instruction, when run on a computer, so that computer executes any one of above-mentioned first aspect or first aspect and realizes
The method.
5th aspect, provides a kind of computer program product comprising instruction, when run on a computer, so that
Computer executes any one of above-mentioned first aspect or first aspect and realizes the method.
The embodiment of the present invention has the advantages that
There is preferable accumulation in terms of algorithm and data since pedestrian identifies again, and bicyclist identifies again, is one
Completely new problem, for trained Finite Samples, transfer learning can identify again that learnt rule carries out reference utilization to pedestrian,
Therefore by carrying out transfer learning to pedestrian's weight identification model, bicyclist's weight identification model is obtained, is identified again using the bicyclist
The feature vector of multiple people's vehicle pictures in model extraction people's vehicle picture library calculates the similarity value between feature vector, and will most
The corresponding at least two people's vehicle picture recognition of the high similarity value is same bicyclist, improves bicyclist and identifies again
Matching degree, realize monitoring to non power driven vehicle illegal activities.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is a kind of exemplary acquisition schematic diagram of people's vehicle picture of the embodiment of the present invention；
Fig. 2 is a kind of flow diagram of bicyclist provided in an embodiment of the present invention recognition methods again；
Fig. 3 is the flow diagram of another bicyclist provided in an embodiment of the present invention recognition methods again；
Fig. 4 is the process schematic provided in an embodiment of the present invention that transfer learning is carried out to pedestrian's weight identification model；
Fig. 5 is a kind of structural schematic diagram of bicyclist's weight identification device provided in an embodiment of the present invention；
Fig. 6 is the structural schematic diagram that a kind of bicyclist provided in an embodiment of the present invention identifies equipment again.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
In order to obtain the more information of people's vehicle, generally, traffic monitoring department can mounted in pairs video camera.As shown in Figure 1,
It is a kind of exemplary acquisition schematic diagram of people's vehicle picture of the embodiment of the present invention, prison is installed in the certain distance above car lane
Equipment pair is controlled, carries out the camera shooting of rear shooting to people's vehicle to the video camera 1 of people's vehicle progress front shooting and one including one
Machine 2.The opposite direction of video camera 1 towards wagon flow is shot, and can capture the illegal non-motor vehicle in car lane traveling and people just
Face, video camera 2 are shot towards wagon flow direction, can capture non-motor vehicle and the back side of people.However, just the lower candid photograph of a monitoring
Dough figurine vehicle monitors lower candid photograph people from back side vehicle with another and matches, i.e. bicyclist identifies again, currently without corresponding technology.It is real
Under the scene of border, manned, object that keeps out the wind etc. can all make the front and back two sides of people's vehicle differ huge, and it is tired that this significantly increases matching
It is difficult.Moreover, can be used for training the data of identification model few.
The embodiment of the present invention provides a kind of bicyclist recognition methods, device and equipment again, by pedestrian's weight identification model
Transfer learning is carried out, bicyclist's weight identification model is obtained, is extracted using bicyclist weight identification model multiple in people's vehicle picture library
The feature vector of people's vehicle picture calculates the similarity value between feature vector, and the highest similarity value is corresponding extremely
Few two people's vehicle picture recognitions are same bicyclist, improve the matching degree that bicyclist identifies again, are realized to non-motor vehicle
The monitoring of illegal activities.
Fig. 2 is a kind of flow diagram of bicyclist provided in an embodiment of the present invention recognition methods again, illustratively, the party
Method can comprise the following steps that
S101, transfer learning is carried out to pedestrian's weight identification model, obtains bicyclist's weight identification model.
Pedestrian identifies that (Person re-identification) is also referred to as pedestrian and identifies again again, is to utilize computer vision skill
Art judges the technology that whether there is specific pedestrian in image or video sequence.The progress that pedestrian identifies again at present is rapid,
On several public data collection, the precision that pedestrian identifies again improves significant.Pedestrian's weight identification model such as residual error network ResNet-
50。
There is preferable accumulation in terms of algorithm and data since pedestrian identifies again, and bicyclist identifies again, is one
Completely new problem, less for trained sample, transfer learning can identify again that learnt rule carries out reference utilization to pedestrian.
Transfer learning is a kind of machine learning method, due to the difference between data set and real data, leads to the training on data set A
Good model performance on real data B is bad.The main method for using transfer learning at present, in the data set for having label
A and without on label data collection B training, finally tested on the test set of data set B.Namely the model developed for task A
As initial point, reuse during for task B development model.Therefore pedestrian's weight identification model is passed through into transfer learning
Change into bicyclist weight identification model be very good selection, can solve in this way bicyclist identify again in Data bottlenecks ask
Topic.
S102, the feature vector that multiple people's vehicle pictures in people's vehicle picture library are extracted using bicyclist weight identification model.
As shown in Figure 1, in order to be supervised to illegal bicyclist, traffic monitoring department mounted in pairs video camera, by this
People's vehicle picture of a little video camera intakes is stored into people's vehicle picture library.Specifically, the people's vehicle picture is the original absorbed from video camera
It detects and is cut out in beginning picture.Multiple people's vehicle pictures are stored in people's vehicle picture library.
After obtaining above-mentioned bicyclist's weight identification model, people's vehicle picture library can be extracted using bicyclist weight identification model
In multiple people's vehicle pictures feature vector, i.e., by people's vehicle picture input bicyclist weight identification model, pass through the complex calculation of model
Obtain the feature vector of people's vehicle picture.In the specific implementation, multiple people's vehicle pictures can be obtained in people's vehicle picture library, can use
Above-mentioned bicyclist's weight identification model extracts the feature vector of multiple people's vehicle pictures one by one or simultaneously.The feature of the people's vehicle picture to
Amount includes appearance, clothing, license plate, the external feature of vehicle etc. of people.
Optionally, the feature vector of a corresponding feature vector of people's vehicle picture or a type.For example, this feature to
Amount is the external feature of bicyclist and vehicle.
Optionally, people's vehicle picture also corresponds to the feature vector of a plurality of feature vector or multiple types.
Multiple similarity values between S103, calculating described eigenvector.
After the feature vector for extracting multiple people's vehicle pictures, multiple similarity values between feature vector are calculated.Specifically
Ground calculates separately the similarity value of any two or multiple feature vectors, between all feature vectors extracted
Multiple similarity values.It can according to need setting and carry out matched feature vector number.
For example, the positive dough figurine vehicle picture and people from back side vehicle picture of bicyclist is stored in people's vehicle picture library respectively, and point
It is not identified, the feature vector of above-mentioned positive dough figurine vehicle picture and people from back side vehicle picture can be extracted respectively, and calculate and appoint
The feature vector of all people from back side vehicle pictures in the feature vector of one positive dough figurine vehicle picture and people's vehicle picture library of extraction it
Between similarity value.
Optionally, be extracted in above-mentioned steps multiple people's vehicle pictures a feature vector or a type feature to
Amount, then calculate the similarity between the feature vector of this multiple people's vehicle picture, obtain multiple similarity values.
Optionally, it is extracted the feature vector of multiple types in above-mentioned steps, then according to the type of feature vector, counts respectively
Calculate the similarity between the feature vector of each type of this multiple people's vehicle picture.It is then possible to according to the feature of each type
The weight of vector calculates the similarity value of the synthesis between the feature vector of each type extracted, as final similar
Angle value.For example, extracting feature vector A1, A2 for people vehicle picture A；For people vehicle picture B, extract feature vector B1,
B2.And feature vector A1, B1 is same type of feature vector, for example, facial eigenvectors of bicyclist；Feature vector A2,
B2 is another type of feature vector, for example, the garment feature vector of bicyclist.This category feature of feature vector A1, B1 is set
The weight of vector is 0.7, and the weight of setting feature vector A2, B2 this kind feature vector is 0.3.Then when calculating feature vector
Similarity value between A1, B1 is 98%, when the similarity value for calculating between feature vector A2, B2 is 80%, obtains people's vehicle
The similarity of the synthesis of the feature vector of picture A, B is 98%*0.7+80%*0.3.
It S104, by the corresponding at least two people's vehicle picture recognition of the highest similarity value is same bicyclist.
After calculating multiple similarity values between features described above vector, highest similarity value is determined, by highest
The corresponding at least two people vehicle picture recognition of similarity value be same bicyclist.In this manner it is possible to which it is more to obtain the bicyclist
Characteristic information, convenient for judge bicyclist whether irregular driving.
For example, the positive dough figurine vehicle picture and people from back side vehicle picture of bicyclist is stored in people's vehicle picture library respectively, and point
It is not identified, all back in people's vehicle picture library of the feature vector and extraction that are extracted any positive dough figurine vehicle picture
After similarity value between the feature vector of dough figurine vehicle picture, it may be determined that similarity value is highest between the positive dough figurine vehicle picture
People from back side vehicle picture, to be same ride by the highest positive dough figurine vehicle picture of this group of similarity value and back side people's vehicle picture recognition
Person.Assuming that bicyclist's irregular driving, but the bicyclist and vehicle can only obtained just according to the positive dough figurine vehicle picture captured
Face, but acquisition will match to and positive dough figurine vehicle picture similarity highest less than the license board information of vehicle by the present processes
People from back side vehicle picture, can determine that the positive dough figurine vehicle picture of the group and people from back side vehicle picture are same bicyclist, then can basis
People from back side vehicle picture gets license board information, to get the more characteristic informations of the bicyclist, rides convenient for judgement
Person whether irregular driving.
A kind of bicyclist recognition methods again provided according to embodiments of the present invention, by being moved to pedestrian's weight identification model
Study is moved, bicyclist's weight identification model is obtained, extracts multiple people Che Tu in people's vehicle picture library using bicyclist weight identification model
The feature vector of piece calculates the similarity value between feature vector, and by the highest similarity value corresponding at least two
People's vehicle picture recognition is same bicyclist, improves the matching degree that bicyclist identifies again, is realized illegal to non power driven vehicle
The monitoring of behavior.
Fig. 3 is a kind of flow diagram of bicyclist provided in an embodiment of the present invention recognition methods again, illustratively, the party
Method can comprise the following steps that
S201, according to carry out front shooting video camera and carry out rear shooting video camera between positional distance,
The storage time of people's vehicle picture in people's vehicle picture library is set.
/ per hour/a large amount of photo will be shot and store per minute since the video camera that is arranged on car lane is daily,
Therefore, in order to save memory space, need to be arranged the storage time of people's vehicle picture in people's vehicle picture library, the people beyond storage time
Vehicle picture deletes outbound.
Generally, carry out front shooting video camera and carry out rear shooting video camera between have certain position away from
From, therefore, obtaining between front shooting photo and reverse side shooting photo has that the regular hour is poor, however, front shooting photo and
It is the highest people's vehicle picture group of similarity that reverse side, which shoots photo, and therefore, it is necessary to front shooting photo and reverse side shooting photo is same
When be retained in people's vehicle picture library, in order to subsequent matching.It therefore, can be according to the video camera and progress for carrying out front shooting
Positional distance between the video camera of rear shooting is arranged the storage time of people's vehicle picture library, or claims holding for people's vehicle picture library
Renew storage duration (time difference of earliest people's vehicle picture in newest storage people's vehicle picture and library).
S202, the image information that multiple people's vehicle pictures are obtained by monitoring device.
As shown in Figure 1, monitoring device is arranged in traffic monitoring department above road, people's vehicle can be absorbed by monitoring device
Picture.Wherein, which includes at least one to the video camera of people's vehicle progress front shooting and one to people Che Jinhang
The video camera of rear shooting, so as to obtain the positive dough figurine vehicle picture and back side people's vehicle picture of bicyclist.
While storing people's vehicle picture, the image information of people's vehicle picture can also be obtained.Wherein, which includes
The candid photograph place of people's vehicle picture and candid photograph time.
S203, according to the candid photograph place and/or the candid photograph time, people's vehicle picture is stored to respective type
People's vehicle picture library.
In the present embodiment, in order to improve matching efficiency, the image credit range of people's vehicle picture library is reduced.Specifically, according to
The corresponding monitoring device in people's vehicle picture source captures place, captures the time or captures place and captures the time to divide people Che Tu
The type of valut.That is people's vehicle picture library only stores the people's vehicle picture for several monitoring devices that some captures place, or
It stores some and captures the people's vehicle picture obtained in time range, or store some some candid photograph time model for capturing place
Enclose people's vehicle picture of interior acquisition.
For example, interval is provided with ten video cameras on a certain lane, wherein video camera A and video camera B is a monitoring
Equipment pair, i.e. video camera A are responsible for shooting people's vehicle front, and video camera B is responsible for shooting people's vehicle back side；Video camera C and video camera D is one
A monitoring device pair, and so on.The video camera A and video camera B photo shot can then be stored to people's vehicle picture library 1, it will
Video camera C and video camera D is stored to people's vehicle picture library 2, etc..In another example video camera A and video camera B are respectively in 8:00~9:
00 shooting obtains multiple people's vehicle pictures, and shoots to obtain multiple people's vehicle pictures respectively in 9:00~10:00, then can be by video camera
A and video camera B is stored in multiple people's vehicle pictures that 8:00~9:00 is shot to people's vehicle picture library 1, by video camera A and camera shooting
Machine B is stored in multiple people's vehicle pictures that 9:00~10:00 is shot to people's vehicle picture library 2, and so on.
S204, the core network parameter for obtaining pedestrian's weight identification model.
S205, the classification number according to bicyclist's training sample are added after the core network of pedestrian weight identification model
Classification layer.
The parameter and the core network parameter of S206, the adjustment classification layer obtain bicyclist's weight identification model.
Above-mentioned steps S204~S206 is to carry out transfer learning to pedestrian's weight identification model, obtains bicyclist and identifies mould again
Type.Its principle is the initial model that pedestrian's weight identification model is considered as to bicyclist's weight identification model, passes through several bicyclists training
The training of sample obtains more accurate bicyclist's weight identification model.As shown in Figure 4 is provided in an embodiment of the present invention to pedestrian
Weight identification model carries out the process schematic of transfer learning, specific training process are as follows: firstly, obtaining a pedestrian trained
Weight identification model.Then, a kind of core network parameter (Resnet-backbone) (residual error network) of the model, and root are obtained
According to the classification number of bicyclist's training sample, addition classification layer (FC) after the core network of pedestrian weight identification model.For example, people
Work sorts out 100,000 bicyclist's training samples, and the training sample for belonging to the same bicyclist is classified as one kind, by above-mentioned 10
Ten thousand bicyclist's training samples are classified as 27000 classes, then according to the classification number of training sample, addition classification layer, in this way after training
The vectors that the output vector of layer of classifying is tieed up for 27000.Finally, first being adjusted according to loss function (softmaxloss) adjusting parameter
The parameter of classification layer, adjusts whole network parameter after losing and stablizing again, and parameter and core network parameter including layer of classifying obtain
To bicyclist's weight identification model.
S207, target person vehicle picture library is obtained.
Classified according to the storage of above-mentioned people's vehicle picture library, the specified people's vehicle picture library to some type of user carry out feature to
The extraction and matching of amount, i.e., it is specified that feature vector is carried out to some candid photograph place and/or some people's vehicle picture for capturing the time
It extracts and matches.Therefore, target person vehicle picture library is obtained, which is the people Che Tu of user's specified type
Valut.
S208, multiple people's vehicle pictures in the target person vehicle picture library are extracted using bicyclist weight identification model
Feature vector.
After obtaining target person vehicle picture library, extracted using bicyclist's weight identification model multiple in the target person vehicle picture library
The feature vector of people's vehicle picture.
Specifically, as shown in following formula 1:
feati=net (xi) ... formula 1
Wherein, xiFor i-th people's vehicle picture, featiFor corresponding people's vehicle feature vector.Net is above-mentioned residual error network
Function.
In specific example, people's vehicle picture may include the positive dough figurine vehicle figure and people from back side Che Tu of each bicyclist, then
S208 includes: the feature vector for extracting the positive dough figurine vehicle figure respectively and the feature vector of people from back side Che Tu.
Multiple similarities between S209, calculating described eigenvector.
After the feature vector for extracting multiple people's vehicle pictures, multiple similarity values between feature vector are calculated.Specifically
Ground calculates separately the similarity value of any two or multiple feature vectors, between all feature vectors extracted
Multiple similarity values.It can according to need setting and carry out matched feature vector number.
In specific example, people's vehicle picture may include the positive dough figurine vehicle figure and people from back side Che Tu of each bicyclist,
S209 is specifically included: the feature vector of the feature vector of the positive dough figurine vehicle picture and people from back side vehicle picture is carried out respectively
Inner product operation obtains the inner product result between the multiple feature vector, wherein the inner product result is as each group of positive dough figurine
Similarity value between the feature vector of vehicle picture and the feature vector of people from back side vehicle picture.Specifically, it can use following
Formula 2 obtains above-mentioned similarity value:
Similarity=InnerProduct (feati,featj) ... formula 2
Wherein, InnerProduct refers to the inner product result between two feature vectors.
It S210, by the corresponding at least two people's vehicle picture recognition of the highest similarity value is same bicyclist.
After calculating multiple similarity values between features described above vector, highest similarity value is determined, by highest
The corresponding at least two people vehicle picture recognition of similarity value be same bicyclist.In this manner it is possible to which it is more to obtain the bicyclist
Characteristic information, convenient for judge bicyclist whether irregular driving.
Specifically, the highest feature vector of similarity can be determined using following formula 3:
Wherein, available so that InnerProduct (feat according to maximum independent variable point set function argmaxi,featj)
Obtain variable point feat corresponding to maximum valueiAnd featj, so that it is determined that highest at least two feature vector of similarity.
After highest at least two feature vector of similarity has been determined, by least two feature vectors corresponding at least two
Personal vehicle picture recognition is same bicyclist.
S211, when people's vehicle picture storage time be more than the setting people's vehicle picture library in people's vehicle picture storage time
When, it is more than people's vehicle picture outbound of the storage time of the setting by storage time.
In order to save memory space, according to the storage time being arranged in above-mentioned steps S201 or claim storage duration,
In people's vehicle picture library that the storage time for judging people's vehicle picture is more than above-mentioned setting when the storage time of people's vehicle picture, it will store
Time is more than people's vehicle picture outbound of the storage time of above-mentioned setting.
A kind of bicyclist recognition methods again provided according to embodiments of the present invention, by being moved to pedestrian's weight identification model
Study is moved, bicyclist's weight identification model is obtained, extracts multiple people Che Tu in people's vehicle picture library using bicyclist weight identification model
The feature vector of piece calculates the similarity value between feature vector, and by the highest similarity value corresponding at least two
People's vehicle picture recognition is same bicyclist, improves the matching degree that bicyclist identifies again, is realized illegal to non power driven vehicle
The monitoring of behavior；And according to capturing the time and/or capturing place, target person vehicle picture library is chosen, target person vehicle picture library is extracted
In people's vehicle picture feature vector carry out similarity mode, to improve the matched efficiency of feature vector of people's vehicle picture；And it is right
The timely outbound of people's vehicle picture of time-out is stored, to save memory space.
The same design of bicyclist in based on the above embodiment recognition methods again, as shown in figure 5, the embodiment of the present invention is also
A kind of bicyclist's weight identification device 100 is provided, which can be applied to the recognition methods again of bicyclist described in above-mentioned Fig. 2, Fig. 3
In.The device 100 includes: study module 11, extraction module 12, computing module 13 and identification module 14, can also include obtaining
Module 15, memory module 16 and setup module 17.It is illustrative:
Study module 11 obtains bicyclist's weight identification model for carrying out transfer learning to pedestrian's weight identification model；
Extraction module 12, for extracting multiple people's vehicle pictures in people's vehicle picture library using bicyclist weight identification model
Feature vector；
Computing module 13, for calculating multiple similarity values between described eigenvector；
Identification module 14, for being same by the corresponding at least two people's vehicle picture recognition of the highest similarity value
One bicyclist.
In one implementation, people's vehicle picture includes the positive dough figurine vehicle figure and people from back side Che Tu of each bicyclist；
The extraction module 12 is specifically used for extracting the feature vector of the positive dough figurine vehicle figure and people from back side vehicle respectively
The feature vector of figure；
The computing module 13 is specifically used for the feature vector of the positive dough figurine vehicle picture and the back side people vehicle picture
Feature vector carry out inner product operation respectively, obtain the inner product result between the multiple feature vector, wherein the inner product knot
Fruit is as the similarity value between the feature vector of each group of positive dough figurine vehicle picture and the feature vector of back side people's vehicle picture.
In another realization, described device further include:
Module 15 is obtained, for obtaining the image information of multiple people's vehicle pictures by monitoring device, wherein the monitoring is set
The standby camera shooting to rear shooting is carried out to people's vehicle to the video camera of people's vehicle progress front shooting and one including at least one
Machine, described image information include the candid photograph place of people's vehicle picture and capture the time；
Memory module 16, for according to the candid photograph place and/or the candid photograph time, by people's vehicle picture store to
People's vehicle picture library of respective type；
The extraction module 12 includes:
First acquisition unit 121, for obtaining target person vehicle picture library, wherein the target person vehicle picture library is described
People's vehicle picture library of user's specified type；
Extraction unit 122, it is multiple in the target person vehicle picture library for being extracted using bicyclist weight identification model
The feature vector of people's vehicle picture.
In another realization, the study module 11 includes:
Second acquisition unit 111, for obtaining the core network parameter of pedestrian's weight identification model；
Adding unit 112, for the classification number according to bicyclist's training sample, in the trunk of pedestrian weight identification model
Addition classification layer after network；
Adjustment unit 113 obtains the bicyclist for adjusting the parameter and the core network parameter of the classification layer
Weight identification model.
In another realization, described device further include:
Setup module 17, for according to the video camera of shooting in front of the progress and the camera shooting of the progress rear shooting
The storage time of people's vehicle picture in people's vehicle picture library is arranged in positional distance between machine；
The memory module 16 is also used to as people in people's vehicle picture library of the storage time of people's vehicle picture more than the setting
It is more than people's vehicle picture outbound of the storage time of the setting by storage time when the storage time of vehicle picture.
It can know again with reference to bicyclist described in above-mentioned Fig. 2, Fig. 3 in relation to above-mentioned modules, unit more detailed description
The associated description of other method obtains, and is not added repeats here.
A kind of bicyclist weight identification device provided according to embodiments of the present invention, by being moved to pedestrian's weight identification model
Study is moved, bicyclist's weight identification model is obtained, extracts multiple people Che Tu in people's vehicle picture library using bicyclist weight identification model
The feature vector of piece calculates the similarity value between feature vector, and by the highest similarity value corresponding at least two
People's vehicle picture recognition is same bicyclist, improves the matching degree that bicyclist identifies again, is realized illegal to non power driven vehicle
The monitoring of behavior；And according to capturing the time and/or capturing place, target person vehicle picture library is chosen, target person vehicle picture library is extracted
In people's vehicle picture feature vector carry out similarity mode, to improve the matched efficiency of feature vector of people's vehicle picture；And it is right
The timely outbound of people's vehicle picture of time-out is stored, to save memory space.
Fig. 6 is the structural schematic diagram that a kind of bicyclist provided in an embodiment of the present invention identifies equipment again, the equipment 200 packet
It includes: including processor 21, may also include input unit 22, output device 23 and memory 24.The input unit 22, output device
23, it is connected with each other between memory 24 and processor 21 by bus.
Memory 24 include but is not limited to be random access memory (random access memory, RAM), read-only deposit
Reservoir (read-only memory, ROM), Erasable Programmable Read Only Memory EPROM (erasable programmable read
Only memory, EPROM) or portable read-only memory (compact disc read-only memory, CD-ROM),
The memory is used for dependent instruction and data.
Input unit 22 is used for output data and/or signal for input data and/or signal and output device 23.
Output device and input unit can be independent device, be also possible to the device of an entirety.
Processor 21 may include be one or more processors, for example including one or more central processing units
(central processing unit, CPU), in the case where processor is a CPU, which can be monokaryon CPU,
It can be multi-core CPU.
Memory 24 is used for the program code and data of storage networking device.
Processor 21 is used to call the program code and data in the memory, executes following steps:
Transfer learning is carried out to pedestrian's weight identification model, obtains bicyclist's weight identification model；
The feature vector of multiple people's vehicle pictures in people's vehicle picture library is extracted using bicyclist weight identification model；
Calculate multiple similarity values between described eigenvector；
It is same bicyclist by the corresponding at least two people's vehicle picture recognition of the highest similarity value.
In one implementation, the processor 21 executes described to pedestrian's weight identification model progress transfer learning, is ridden
The step of passerby's weight identification model, comprising:
Obtain the core network parameter of pedestrian's weight identification model；
According to the classification number of bicyclist's training sample, classification is added after the core network of pedestrian weight identification model
Layer；
The parameter and the core network parameter for adjusting the classification layer obtain bicyclist's weight identification model.
In another realization, people's vehicle picture includes the positive dough figurine vehicle figure and people from back side Che Tu of each bicyclist, institute
It states processor 21 and executes the feature for extracting multiple people's vehicle pictures in people's vehicle picture library using bicyclist weight identification model
The step of vector, comprising:
The feature vector of the positive dough figurine vehicle figure and the feature vector of people from back side Che Tu are extracted respectively；
The processor 21 executes the step of multiple similarity values between the calculating described eigenvector, comprising:
The feature vector of the feature vector of the positive dough figurine vehicle picture and people from back side vehicle picture is subjected to inner product respectively
Operation obtains the inner product result between the multiple feature vector, wherein the inner product result is as each group of positive dough figurine vehicle figure
Similarity value between the feature vector of piece and the feature vector of people from back side vehicle picture.
In another realization, the processor 21 is also executed the following steps:
The image information of multiple people's vehicle pictures is obtained by monitoring device, wherein the monitoring device is to including at least one
A video camera and a video camera to the progress rear shooting of people's vehicle that front shooting is carried out to people's vehicle, described image information
Candid photograph place and candid photograph time including people's vehicle picture；
According to the candid photograph place and/or the candid photograph time, people's vehicle picture is stored to people's vehicle of respective type
Picture library；
The processor 21 executes use bicyclist weight identification model and extracts multiple people in people's vehicle picture library
The step of feature vector of vehicle picture, comprising:
Obtain target person vehicle picture library, wherein the target person vehicle picture library is the people Che Tu of user's specified type
Valut；
The spy of multiple people's vehicle pictures in the target person vehicle picture library is extracted using bicyclist weight identification model
Levy vector.
In another realization, the processor 21 is also executed the following steps:
According to the position between the video camera of shooting in front of the progress and the video camera for carrying out rear shooting away from
From the storage time of people's vehicle picture in people's vehicle picture library is arranged；
It, will when the storage time of people's vehicle picture in people's vehicle picture library that the storage time of people's vehicle picture is more than the setting
Storage time is more than people's vehicle picture outbound of the storage time of the setting.
It is understood that Fig. 5, which illustrate only bicyclist, identifies that simplifying for equipment is designed again.In practical applications, electric
Sub- equipment can also separately include necessary other elements, including but not limited to any number of input/output device, processing
Device, controller, memory etc., and all electronic equipments that the embodiment of the present application may be implemented all the protection scope of the application it
It is interior.
A kind of bicyclist provided according to embodiments of the present invention identifies equipment again, by moving to pedestrian's weight identification model
Study is moved, bicyclist's weight identification model is obtained, extracts multiple people Che Tu in people's vehicle picture library using bicyclist weight identification model
The feature vector of piece calculates the similarity value between feature vector, and by the highest similarity value corresponding at least two
People's vehicle picture recognition is same bicyclist, improves the matching degree that bicyclist identifies again, is realized illegal to non power driven vehicle
The monitoring of behavior；And according to capturing the time and/or capturing place, target person vehicle picture library is chosen, target person vehicle picture library is extracted
In people's vehicle picture feature vector carry out similarity mode, to improve the matched efficiency of feature vector of people's vehicle picture；And it is right
The timely outbound of people's vehicle picture of time-out is stored, to save memory space.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description,
The specific work process of device and unit, can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
In several embodiments provided herein, it should be understood that disclosed systems, devices and methods, it can be with
It realizes by another way.For example, the division of the unit, only a kind of logical function partition, can have in actual implementation
Other division mode, for example, multiple units or components can be combined or can be integrated into another system or some features
It can ignore, or not execute.Shown or discussed mutual coupling or direct-coupling or communication connection can be logical
Some interfaces are crossed, the indirect coupling or communication connection of device or unit can be electrical property, mechanical or other forms.
Unit may or may not be physically separated as illustrated by the separation member, shown as a unit
Component may or may not be physical unit, it can and it is in one place, or may be distributed over multiple networks
On unit.It can some or all of the units may be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
In the above-described embodiments, can come wholly or partly by software, hardware, firmware or any combination thereof real
It is existing.When implemented in software, it can entirely or partly realize in the form of a computer program product.The computer program produces
Product include one or more computer instructions.It is all or part of when loading and execute on computers the computer program instructions
Ground generates the process or function according to the embodiment of the present application.The computer can be general purpose computer, special purpose computer, computer
Network or other programmable devices.The computer instruction may be stored in a computer readable storage medium, or by being somebody's turn to do
Computer readable storage medium is transmitted.The computer instruction can be from a web-site, computer, server or data
Center passes through wired (such as coaxial cable, optical fiber, Digital Subscriber Line (digital subscriber line, DSL)) or wireless
(such as infrared, wireless, microwave etc.) mode is transmitted to another web-site, computer, server or data center.It should
Computer readable storage medium can be any usable medium that computer can access or include one or more available
The data storage devices such as medium integrated server, data center.The usable medium can be read-only memory (read-only
Memory, ROM) or random access memory (random access memory, RAM) or magnetic medium, for example, floppy disk,
Hard disk, tape, magnetic disk or optical medium, for example, digital versatile disc (digital versatile disc, DVD) or half
Conductive medium, for example, solid state hard disk (solid state disk, SSD) etc..
Claims (10)
1. a kind of bicyclist recognition methods again characterized by comprising
Transfer learning is carried out to pedestrian's weight identification model, obtains bicyclist's weight identification model；
The feature vector of multiple people's vehicle pictures in people's vehicle picture library is extracted using bicyclist weight identification model；
Calculate multiple similarity values between described eigenvector；
It is same bicyclist by the corresponding at least two people's vehicle picture recognition of the highest similarity value.
2. being obtained the method according to claim 1, wherein described carry out transfer learning to pedestrian's weight identification model
To bicyclist's weight identification model, comprising:
Obtain the core network parameter of pedestrian's weight identification model；
According to the classification number of bicyclist's training sample, the addition classification layer after the core network of pedestrian weight identification model；
The parameter and the core network parameter for adjusting the classification layer obtain bicyclist's weight identification model.
3. the method according to claim 1, wherein people's vehicle picture includes the positive dough figurine vehicle of each bicyclist
Figure and people from back side Che Tu, the feature that multiple people's vehicle pictures in people's vehicle picture library are extracted using bicyclist weight identification model
Vector, comprising:
The feature vector of the positive dough figurine vehicle figure and the feature vector of people from back side Che Tu are extracted respectively；
Multiple similarity values between the calculating described eigenvector, comprising:
The feature vector of the feature vector of the positive dough figurine vehicle picture and people from back side vehicle picture is subjected to inner product operation respectively,
Obtain the inner product result between the multiple feature vector, wherein the inner product result is as each group of positive dough figurine vehicle picture
Similarity value between feature vector and the feature vector of people from back side vehicle picture.
4. method described in any one of claim 1 to 3, which is characterized in that the method also includes:
The image information of multiple people's vehicle pictures is obtained by monitoring device, wherein the monitoring device includes at least one to people
Vehicle carries out the video camera and a video camera to the progress rear shooting of people's vehicle of front shooting, and described image information includes institute
It states the candid photograph place of people's vehicle picture and captures the time；
According to the candid photograph place and/or the candid photograph time, people's vehicle picture is stored to people's vehicle picture of respective type
Library；
The feature vector that multiple people's vehicle pictures in people's vehicle picture library are extracted using bicyclist weight identification model, packet
It includes:
Obtain target person vehicle picture library, wherein the target person vehicle picture library is people's vehicle picture library of user's specified type；
Using bicyclist weight identification model extract the features of multiple people's vehicle pictures in the target person vehicle picture library to
Amount.
5. according to the method described in claim 4, it is characterized in that, the method also includes:
According to the positional distance between the video camera of shooting in front of the progress and the video camera for carrying out rear shooting, if
Set the storage time of people's vehicle picture in people's vehicle picture library；
When the storage time of people's vehicle picture in people's vehicle picture library that the storage time of people's vehicle picture is more than the setting, will store
Time is more than people's vehicle picture outbound of the storage time of the setting.
6. a kind of bicyclist's weight identification device characterized by comprising
Study module obtains bicyclist's weight identification model for carrying out transfer learning to pedestrian's weight identification model；
Extraction module, for using the bicyclist weight identification model extract people's vehicle picture library in multiple people's vehicle pictures feature to
Amount；
Computing module, for calculating multiple similarity values between described eigenvector；
Identification module, for being same ride by the corresponding at least two people's vehicle picture recognition of the highest similarity value
Person.
7. device according to claim 6, which is characterized in that people's vehicle picture includes the positive dough figurine vehicle of each bicyclist
Figure and people from back side Che Tu；
The extraction module is specifically used for extracting the feature vector of the positive dough figurine vehicle figure and the spy of people from back side Che Tu respectively
Levy vector；
The computing module is specifically used for the feature of the feature vector of the positive dough figurine vehicle picture and the back side people vehicle picture
Vector carries out inner product operation respectively, obtains the inner product result between the multiple feature vector, wherein the inner product result conduct
Similarity value between the feature vector of each group of positive dough figurine vehicle picture and the feature vector of people from back side vehicle picture.
8. device according to claim 6 or 7, which is characterized in that described device further include:
Module is obtained, for obtaining the image information of multiple people's vehicle pictures by monitoring device, wherein the monitoring device is at least
The video camera and a video camera to the progress rear shooting of people's vehicle of front shooting, the figure are carried out to people's vehicle including one
As information includes the candid photograph place of people's vehicle picture and is captured the time；
Memory module, for according to the candid photograph place and/or the candid photograph time, people's vehicle picture to be stored to respective class
People's vehicle picture library of type；
The extraction module includes:
First acquisition unit, for obtaining target person vehicle picture library, wherein the target person vehicle picture library is specified for the user
People's vehicle picture library of type；
Extraction unit, for extracting multiple people's vehicles in the target person vehicle picture library using bicyclist weight identification model
The feature vector of picture.
9. a kind of bicyclist identifies equipment again, which is characterized in that including processor, input equipment, output equipment and memory, institute
Memory is stated for storing computer program, the computer program includes program instruction, and the processor is configured for adjusting
It is instructed with described program, executes the method as described in any claim in claim 1 to 5.
10. a kind of computer readable storage medium, it is stored with instruction in the computer readable storage medium, when it is in computer
When upper operation, so that computer executes the method as described in any claim in claim 1 to 5.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910550548.6A CN110427814A (en) | 2019-06-24 | 2019-06-24 | A kind of bicyclist recognition methods, device and equipment again |
| PCT/CN2019/121517 WO2020258714A1 (en) | 2019-06-24 | 2019-11-28 | Rider re-identification method, apparatus and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910550548.6A CN110427814A (en) | 2019-06-24 | 2019-06-24 | A kind of bicyclist recognition methods, device and equipment again |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110427814A true CN110427814A (en) | 2019-11-08 |
Family
ID=68409506
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910550548.6A Pending CN110427814A (en) | 2019-06-24 | 2019-06-24 | A kind of bicyclist recognition methods, device and equipment again |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN110427814A (en) |
| WO (1) | WO2020258714A1 (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582277A (en) * | 2020-06-15 | 2020-08-25 | 深圳天海宸光科技有限公司 | License plate recognition system and method based on transfer learning |
| WO2020258714A1 (en) * | 2019-06-24 | 2020-12-30 | 深圳云天励飞技术有限公司 | Rider re-identification method, apparatus and device |
| CN112464922A (en) * | 2021-02-02 | 2021-03-09 | 长沙海信智能系统研究院有限公司 | Human-vehicle weight recognition and model training method, device, equipment and storage medium thereof |
| CN113129597A (en) * | 2019-12-31 | 2021-07-16 | 深圳云天励飞技术有限公司 | Method and device for identifying illegal vehicles on motor vehicle lane |
| CN113627352A (en) * | 2021-08-12 | 2021-11-09 | 塔里木大学 | Pedestrian re-identification method and system |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114821629A (en) * | 2021-01-27 | 2022-07-29 | 天津大学 | Pedestrian re-identification method for performing cross image feature fusion based on neural network parallel training architecture |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944017A (en) * | 2017-12-11 | 2018-04-20 | 浙江捷尚视觉科技股份有限公司 | The search method of non-motor vehicle in a kind of video |
| CN108021933A (en) * | 2017-11-23 | 2018-05-11 | 深圳市华尊科技股份有限公司 | Neural network recognization model and recognition methods |
| CN109344842A (en) * | 2018-08-15 | 2019-02-15 | 天津大学 | A kind of pedestrian's recognition methods again based on semantic region expression |
| CN109446898A (en) * | 2018-09-20 | 2019-03-08 | 暨南大学 | A kind of recognition methods again of the pedestrian based on transfer learning and Fusion Features |
| CN109657552A (en) * | 2018-11-16 | 2019-04-19 | 北京邮电大学 | The vehicle type recognition device being cold-started across scene and method are realized based on transfer learning |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106778804B (en) * | 2016-11-18 | 2020-10-20 | 天津大学 | Zero sample image classification method based on class attribute transfer learning |
| CN109002761A (en) * | 2018-06-13 | 2018-12-14 | 中山大学新华学院 | A kind of pedestrian's weight identification monitoring system based on depth convolutional neural networks |
| CN109117823A (en) * | 2018-08-31 | 2019-01-01 | 常州大学 | A kind of across the scene pedestrian based on multilayer neural network knows method for distinguishing again |
| CN110427814A (en) * | 2019-06-24 | 2019-11-08 | 深圳云天励飞技术有限公司 | A kind of bicyclist recognition methods, device and equipment again |
-
2019
- 2019-06-24 CN CN201910550548.6A patent/CN110427814A/en active Pending
- 2019-11-28 WO PCT/CN2019/121517 patent/WO2020258714A1/en active Application Filing
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108021933A (en) * | 2017-11-23 | 2018-05-11 | 深圳市华尊科技股份有限公司 | Neural network recognization model and recognition methods |
| CN107944017A (en) * | 2017-12-11 | 2018-04-20 | 浙江捷尚视觉科技股份有限公司 | The search method of non-motor vehicle in a kind of video |
| CN109344842A (en) * | 2018-08-15 | 2019-02-15 | 天津大学 | A kind of pedestrian's recognition methods again based on semantic region expression |
| CN109446898A (en) * | 2018-09-20 | 2019-03-08 | 暨南大学 | A kind of recognition methods again of the pedestrian based on transfer learning and Fusion Features |
| CN109657552A (en) * | 2018-11-16 | 2019-04-19 | 北京邮电大学 | The vehicle type recognition device being cold-started across scene and method are realized based on transfer learning |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020258714A1 (en) * | 2019-06-24 | 2020-12-30 | 深圳云天励飞技术有限公司 | Rider re-identification method, apparatus and device |
| CN113129597A (en) * | 2019-12-31 | 2021-07-16 | 深圳云天励飞技术有限公司 | Method and device for identifying illegal vehicles on motor vehicle lane |
| CN111582277A (en) * | 2020-06-15 | 2020-08-25 | 深圳天海宸光科技有限公司 | License plate recognition system and method based on transfer learning |
| CN112464922A (en) * | 2021-02-02 | 2021-03-09 | 长沙海信智能系统研究院有限公司 | Human-vehicle weight recognition and model training method, device, equipment and storage medium thereof |
| CN112464922B (en) * | 2021-02-02 | 2021-05-28 | 长沙海信智能系统研究院有限公司 | Human-vehicle weight recognition and model training method, device, equipment and storage medium thereof |
| CN113627352A (en) * | 2021-08-12 | 2021-11-09 | 塔里木大学 | Pedestrian re-identification method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2020258714A1 (en) | 2020-12-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110427814A (en) | A kind of bicyclist recognition methods, device and equipment again | |
| CN108229335A (en) | It is associated with face identification method and device, electronic equipment, storage medium, program | |
| WO2021135879A1 (en) | Vehicle data monitoring method and apparatus, computer device, and storage medium | |
| WO2019127273A1 (en) | Multi-person face detection method, apparatus, server, system, and storage medium | |
| CN109819208A (en) | A kind of dense population security monitoring management method based on artificial intelligence dynamic monitoring | |
| CN105160318A (en) | Facial expression based lie detection method and system | |
| CN108920995A (en) | Intelligent security guard video monitoring method and its system and monitor terminal | |
| CN110399835B (en) | Analysis method, device and system for personnel residence time | |
| CN108446681B (en) | Pedestrian analysis method, device, terminal and storage medium | |
| CN109214263A (en) | A kind of face identification method based on feature multiplexing | |
| KR101493009B1 (en) | Method for front and rear vehicle license plate recognition and system thereof | |
| CN110334231A (en) | A kind of information processing method and device, storage medium | |
| CN106471440A (en) | Eye tracking based on efficient forest sensing | |
| Ali et al. | Autonomous road surveillance system: A proposed model for vehicle detection and traffic signal control | |
| CN111753601B (en) | Image processing method, device and storage medium | |
| CN111177469A (en) | Face retrieval method and face retrieval device | |
| CN110517500A (en) | A kind of people's vehicle association process method and device | |
| CN116186770A (en) | Image desensitizing method, device, electronic equipment and storage medium | |
| CN112232178A (en) | Region footfall point judgment method, system, device and medium based on portrait gathering | |
| CN110188602A (en) | Face identification method and device in video | |
| CN112651398A (en) | Vehicle snapshot control method and device and computer readable storage medium | |
| CN116052059A (en) | Traffic illegal behavior detection method, device and system | |
| CN106682590B (en) | Processing method of monitoring service and server | |
| CN111783674A (en) | Face recognition method and system based on AR glasses | |
| CN110324589A (en) | A kind of monitoring system and method for tourist attraction |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 518000 1st floor, building 17, Shenzhen Dayun software Town, 8288 Longgang Avenue, Yuanshan street, Longgang District, Shenzhen City, Guangdong Province Applicant after: Shenzhen Yuntian lifeI Technology Co.,Ltd. Address before: 518000 1st floor, building 17, Shenzhen Dayun software Town, 8288 Longgang Avenue, Yuanshan street, Longgang District, Shenzhen City, Guangdong Province Applicant before: SHENZHEN INTELLIFUSION TECHNOLOGIES Co.,Ltd. |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191108 |