CN110516640A - It is a kind of to combine the vehicle indicated discrimination method again based on feature pyramid - Google Patents

It is a kind of to combine the vehicle indicated discrimination method again based on feature pyramid Download PDF

Info

Publication number
CN110516640A
CN110516640A CN201910818186.4A CN201910818186A CN110516640A CN 110516640 A CN110516640 A CN 110516640A CN 201910818186 A CN201910818186 A CN 201910818186A CN 110516640 A CN110516640 A CN 110516640A
Authority
CN
China
Prior art keywords
vehicle
feature
discrimination method
image
training
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN201910818186.4A
Other languages
Chinese (zh)
Other versions
CN110516640B (en
Inventor
曾焕强
林向伟
朱建清
曹九稳
陈婧
张云
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huaqiao University
Original Assignee
Huaqiao University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Huaqiao University filed Critical Huaqiao University
Priority to CN201910818186.4A priority Critical patent/CN110516640B/en
Publication of CN110516640A publication Critical patent/CN110516640A/en
Application granted granted Critical
Publication of CN110516640B publication Critical patent/CN110516640B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
    • G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/08—Detecting or categorising vehicles
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The present invention relates to a kind of vehicle discrimination methods again indicated based on feature pyramid joint, comprising: construction feature pyramid designs the pyramidal joint representation method of multi-stage characteristics, and allowable loss function carries out learning distance metric to the image of input and comparison learns.The present invention special consideration should be given to vehicle recognize again in difficult point, i.e., the vehicle image resolution ratio captured by the camera of different distance is different, can efficiently solve vehicle recognize again in the not high problem of the vehicle identification precision that generates for above-mentioned difficult point.

Description

It is a kind of to combine the vehicle indicated discrimination method again based on feature pyramid
Technical field
The present invention relates to computer visions and area of pattern recognition, more particularly to a kind of feature pyramid joint that is based on to indicate Vehicle discrimination method again.
Background technique
Vehicle recognizes again to be intended to from retrieving the current queries vehicle occurred in monitor video in monitor video, Its motion profile is tracked, there is very important practical application value.Vehicle identifies that application scenarios are very extensive again, such as vehicle Tracking, vehicle location, criminal's detection etc..
Change in illumination variation, viewpoint, block, in the factors such as change resolution, change resolution (i.e. by difference away from From the collected vehicle image resolution ratio of video camera it is different) be difficult point generally existing during vehicle recognizes again.Existing vehicle is again Discrimination method has that accuracy of identification is poor for above-mentioned resolution ratio different vehicle image.Therefore, how vehicle is fully considered Discrimination method becomes vehicle and recognizes the challenging research hotspot in one, field again the efficient vehicle of image resolution ratio again.
Summary of the invention
Identification technique is not high for the vehicle image accuracy of identification of different resolution again asks for existing vehicle by the present invention Topic is proposed a kind of vehicle discrimination method again indicated based on feature pyramid joint, passes through construction feature pyramid, and design is multistage The pyramidal joint representation method of feature, and corresponding loss function is combined, distance metric is carried out to the vehicle image of input It practises, effectively improves the accuracy of identification that vehicle recognizes again.
Technical solution used by its technical problem of solution of the present invention is:
A kind of vehicle discrimination method again indicated based on feature pyramid joint, including training process and identification process again, Steps are as follows:
Training process S1: the different vehicle image of resolution ratio, which is input to feature pyramid joint, to be indicated to instruct in network model Practice, until model is restrained, to obtain network model M；
Identification process S2 again: network model M is obtained using training in S1 and extracts vehicle image and candidate storehouse to be checked respectively In each image feature vector, the similarity of image in vehicle image to be checked and candidate storehouse, root are calculated according to feature vector Vehicle ID prediction is carried out according to the size of similarity, and exports recognition result.
Preferably, the step of training process S1 is as follows:
Step S11): the vehicle image of different resolution is input in basic network and extracts vehicle characteristics；
Step S12): vehicle characteristics extracted in S11) are classified, extracting at different levels includes different resolution and language The characteristic block of adopted information strength；
Step S13): to S12) in extracted characteristic blocks at different levels by convolution operation carry out dimensionality reduction, respectively obtain resolution The identical characteristic block x of rate difference dimension1, x2, x3, x4；
Step S14): to S13) in characteristic block after dimensionality reduction, from network bottom layer by up-sampling operation gradually to network top Layer is rebuild, and the characteristic block f of four reconstructions is obtained1, f2, f3, f4；
Step S15): to S14) obtained in four reconstruction features block f1, f2, f3, f4Two are respectively obtained by pondization operation Dimensional feature vector, and carry out splicing and realize that joint indicates, obtain union feature vector fC；
Step S16): union feature vector obtained in S15) is respectively fed to learning distance metric and base categories module In, it calculates corresponding penalty values and addition in certain proportion obtains combination learning penalty values, calculated using error back propagation Method, which carrys out training characteristics pyramid joint, indicates network model；
Step S17): repeat step S11) to step S16), combining until features described above pyramid indicates that network model is received It holds back.
Preferably, step S11) in basic network be made of ResNet-50.
Preferably, the basic network includes a convolutional layer, pond layer and four convolution blocks, and each convolution block includes not With the residual block of quantity.
Preferably, step S12) in characteristic block x1, x2, x3, x4Output Size be respectively (256 × 64 × 64), (512 × 32 × 32), (1024 × 16 × 16), (2048 × 16 × 16).
Preferably, step S14) in rebuild characteristic block f1, f2, f3, f4Output Size be respectively (256 × 16 × 16), (256 × 16 × 16), (256 × 32 × 32), (256 × 64 × 64).
Preferably, step S16) in, the loss formula of learning distance metric are as follows:
Wherein, [x]+=max (x, 0), Ia,iFor the sample randomly selected in training set, Ip,iFor with Ia,iBelong to same A kind of sample, In,iFor with Ia,iInhomogeneous sample, α are a constant, indicate the smallest difference, M between positive negative sample pair Indicate the number of positive/negative sample group；The target of the loss function is to ensure that negative sample to f (Ia,i) and f (In,i) between feature Distance is greater than positive sample to f (Ia,i) and f (In,iThe distance between).
Preferably, step S16) in, the loss formula of base categories module are as follows:
Wherein, 1 () was indicative function, yk∈ 1,2,3 ..., and C } it is vehicle image xkCorresponding ID, K and C generation respectively The quantity of table training sample and ID, H=[H1,H2,...,HC] it is the mapping matrix for predicting vehicle ID.
Preferably, step S16) in, the combination learning penalty values are as follows:
Loss=LTriplet+λ·LSoftmax
Wherein, λ is adjustment LTripletAnd LSoftmaxParameter, being worth is 0.5.
Preferably, then the step of identification process S2, is as follows:
Step S21): using the network model M that training obtains in S1, the feature vector of vehicle image to be checked is extracted respectively And in candidate storehouse each image feature vector；
Step S22): each image in the feature vector and candidate storehouse of vehicle image to be checked is calculated by Euclidean distance Similarity between feature vector；
Step S23): descending sort is carried out according to obtained similarity size, and exports recognition result.
Beneficial effects of the present invention are as follows:
The present invention constructs a kind of vehicle identification network again indicated based on feature pyramid joint, passes through construction feature gold Word tower, design multi-stage characteristics are combined representation method, are optimized in conjunction with vehicle characteristics more abundant to the network model, so that The vehicle indicated based on feature pyramid joint that training obtains recognizes vehicle of the model with degree of precision identification capability again again, Especially for the lower vehicle image of resolution ratio；It is instructed in identification process again using based on feature pyramid joint representation method The network model got treats the vehicle pictures in enquiring vehicle picture and candidate storehouse and carries out feature extraction and Euclidean distance It calculates, to obtain the similarity between the vehicle image in vehicle and candidate storehouse to be checked, realizes that vehicle recognizes again.This method It can be widely used in intelligent video monitoring scene, such as vehicle location, track of vehicle prediction, criminal's tracking etc..
Detailed description of the invention
Fig. 1 is the block schematic illustration of the vehicle discrimination method again indicated the present invention is based on feature pyramid joint.
Specific embodiment
The present invention is further described in detail with reference to the accompanying drawings and embodiments.
Shown in Figure 1, the present invention is a kind of to combine the vehicle indicated discrimination method again based on feature pyramid, is based on feature The vehicle that pyramid joint indicates recognizes model again, which includes characteristic extracting module and learning distance metric module, including Training process and again identification process, the specific steps are as follows:
Training process S1: the different vehicle image of resolution ratio, which is input to feature pyramid joint, to be indicated to instruct in network model Practice, until model is restrained, to obtain network model M；
The step of training process S1, is as follows:
Step S11): the vehicle image of different resolution is input in basic network and extracts vehicle characteristics.
Step S12): vehicle characteristics extracted in S11) are classified, extracting at different levels includes different resolution and language The characteristic block of adopted information strength.
Step S13): to S12) in extracted characteristic blocks at different levels by convolution operation carry out dimensionality reduction, respectively obtain resolution The identical characteristic block x of rate difference dimension1, x2, x3, x4。
Step S14): to S13) in characteristic block after dimensionality reduction, from network bottom layer by up-sampling operation gradually to network top Layer is rebuild, and the characteristic block f of four reconstructions is obtained1, f2, f3, f4。
Step S15): to S14) obtained in four reconstruction features block f1, f2, f3, f4Two are respectively obtained by pondization operation Dimensional feature vector, and carry out splicing and realize that joint indicates, obtain union feature vector fC。
Step S16): union feature vector obtained in S15) is respectively fed to learning distance metric and base categories module In, it calculates corresponding penalty values and addition in certain proportion obtains combination learning penalty values, calculated using error back propagation Method, which carrys out training characteristics pyramid joint, indicates network model.
Step S17): repeat step S11) to step S16), combining until features described above pyramid indicates that network model is received It holds back.
Further, step S11) in basic network be made of ResNet-50.
Further, the basic network includes a convolutional layer, pond layer and four convolution blocks, and each convolution block includes The identical residual block of different number.
Further, step S12) in characteristic block x1, x2, x3, x4Output Size be respectively (256 × 64 × 64), (512 × 32 × 32), (1024 × 16 × 16), (2048 × 16 × 16).
Further, step S14) in rebuild characteristic block f1, f2, f3, f4Output Size be respectively (256 × 16 × 16), (256 × 16 × 16), (256 × 32 × 32), (256 × 64 × 64).
Further, step S16) in, the loss formula of learning distance metric are as follows:
Wherein, [x]+=max (x, 0), Ia,iFor the sample randomly selected in training set, Ip,iFor with Ia,iBelong to same A kind of sample, In,iFor with Ia,iInhomogeneous sample, α are a constant, indicate the smallest difference, M between positive negative sample pair Indicate the number of positive/negative sample group；The target of the loss function is to ensure that negative sample to f (Ia,i) and f (In,i) between feature Distance is greater than positive sample to f (Ia,i) and f (In,iThe distance between).
Further, step S16) in, the loss formula of base categories module are as follows:
Wherein, 1 () was indicative function, yk∈ 1,2,3 ..., and C } it is vehicle image xkCorresponding ID, K and C generation respectively The quantity of table training sample and ID, H=[H1,H2,...,HC] it is the mapping matrix for predicting vehicle ID.
Further, step S16), combination learning penalty values are as follows:
Loss=LTriplet+λ·LSoftmax
Wherein, λ is adjustment LTripletAnd LSoftmaxParameter, being worth is 0.5.
Identification process S2 again: network model M is obtained using training in S1 and extracts vehicle image and candidate storehouse to be checked respectively In each image feature vector, according to feature vector calculate vehicle image to be checked in candidate storehouse vehicle image it is similar Degree carries out vehicle ID prediction according to the size of similarity, and exports recognition result.
The step of identification process S2 is as follows again:
Step S21): as shown in Figure 1, using the network model M that training obtains in S1, it is corresponding to extract vehicle image to be checked Feature, be denoted as feature vector Q；Each corresponding feature of image in candidate storehouse is extracted, feature vector 1, feature vector are denoted as 2 ..., feature vector N.
Step S22): calculate the feature vector of each vehicle image in the feature vector and candidate storehouse of vehicle image to be checked Between Euclidean distance, Euclidean distance it is smaller indicate two images similarity it is bigger, to get vehicle image to be checked With the similarity relationship of vehicle image each in candidate storehouse.
Step S23): descending sort is carried out according to obtained similarity size, and exports recognition result, is looked into according to vehicle The optimal index of inquiry, if the image and image to be checked vehicle ID having the same of output sort result first, illustrate this Successful inquiring, otherwise, inquiry failure.
Above-described embodiment is intended merely to illustrate the present invention, and is not used as limitation of the invention.As long as according to this hair Bright technical spirit is changed above-described embodiment, modification etc. will all be fallen in the scope of the claims of the invention.

Claims (9)

1. a kind of combine the vehicle indicated discrimination method again based on feature pyramid, which is characterized in that including training process and again Identification process, steps are as follows:
Training process S1: the different vehicle image of resolution ratio, which is input to feature pyramid joint, indicates training in network model, Until model is restrained, to obtain network model M；
Identification process S2 again: using training in S1 obtain network model M extract respectively it is every in vehicle image and candidate storehouse to be checked The feature vector of a image calculates the similarity of vehicle image in vehicle image to be checked and candidate storehouse, root according to feature vector Vehicle ID prediction is carried out according to the size of similarity, and exports recognition result.
2. discrimination method, feature exist a kind of vehicle indicated based on feature pyramid joint according to claim 1 again It is as follows in, the training process S1 the step of:
Step S11): the vehicle image of different resolution is input in basic network and extracts vehicle characteristics；
Step S12): vehicle characteristics extracted in S11) are classified, are extracted at different levels comprising different resolution and semantic letter Cease the characteristic block of intensity；
Step S13): to S12) in extracted characteristic blocks at different levels by convolution operation carry out dimensionality reduction, respectively obtain resolution ratio not The identical characteristic block x with dimension1, x2, x3, x4；
Step S14): to S13) in characteristic block after dimensionality reduction, from network bottom layer by up-sampling operation gradually to network top weight It builds, obtains the characteristic block f of four reconstructions1, f2, f3, f4；
Step S15): to S14) obtained in four reconstruction features block f1, f2, f3, f4Two-dimentional spy is respectively obtained by pondization operation Vector is levied, and carries out splicing and realizes that joint indicates, obtains union feature vector fC；
Step S16): union feature vector obtained in S15) is respectively fed in learning distance metric and base categories module, Calculate corresponding penalty values and in certain proportion be added obtain combination learning penalty values, using error backpropagation algorithm come Training characteristics pyramid joint indicates network model；
Step S17): repeat step S11) to step S16), combining until features described above pyramid indicates network model convergence.
3. discrimination method, feature exist a kind of vehicle indicated based on feature pyramid joint according to claim 2 again Basic network is made of ResNet-50 in step S11), including a convolutional layer, pond layer and four convolution blocks, Mei Gejuan Block includes the residual block of different number.
4. discrimination method, feature exist a kind of vehicle indicated based on feature pyramid joint according to claim 2 again The characteristic block x in step S12)1, x2, x3, x4Output Size be respectively (256 × 64 × 64), (512 × 32 × 32), (1024 × 16 × 16), (2048 × 16 × 16).
5. discrimination method, feature exist a kind of vehicle indicated based on feature pyramid joint according to claim 2 again The characteristic block f rebuild in step S14)1, f2, f3, f4Output Size be respectively (256 × 16 × 16), (256 × 16 × 16), (256 × 32 × 32), (256 × 64 × 64).
6. discrimination method, feature exist a kind of vehicle indicated based on feature pyramid joint according to claim 2 again In step S16), the loss formula of learning distance metric are as follows:
Wherein, [x]+=max (x, 0), Ia,iFor the sample randomly selected in training set, Ip,iFor with Ia,iBelong to of a sort Sample, In,iFor with Ia,iInhomogeneous sample, α are a constant, indicate the smallest difference between positive negative sample pair, M indicate just/ The number of negative sample group；The target of the loss function is to ensure that negative sample to f (Ia,i) and f (In,i) between characteristic distance be greater than Positive sample is to f (Ia,i) and f (In,iThe distance between).
7. discrimination method, feature exist a kind of vehicle indicated based on feature pyramid joint according to claim 2 again In step S16), the loss formula of base categories module are as follows:
Wherein, 1 () was indicative function, yk∈ 1,2,3 ..., and C } it is vehicle image xkCorresponding ID, K and C respectively represent instruction Practice the quantity of sample and ID, H=[H1,H2,...,HC] it is the mapping matrix for predicting vehicle ID.
8. discrimination method, feature exist a kind of vehicle indicated based on feature pyramid joint according to claim 2 again In step S16), the combination learning penalty values are as follows:
Loss=LTriplet+λ·LSoftmax
Wherein, λ is adjustment LTripletAnd LSoftmaxParameter, being worth is 0.5.
9. discrimination method, feature exist a kind of vehicle indicated based on feature pyramid joint according to claim 1 again In, then the step of identification process S2 is as follows:
Step S21): using the basic network model M that training obtains in S1, the feature vector of vehicle image to be checked is extracted respectively And in candidate storehouse each image feature vector；
Step S22): the feature of each image in the feature vector and candidate storehouse of vehicle image to be checked is calculated by Euclidean distance Similarity between vector；
Step S23): descending sort is carried out according to obtained similarity size, and exports recognition result.
CN201910818186.4A 2019-08-30 2019-08-30 Vehicle re-identification method based on feature pyramid joint representation Active CN110516640B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910818186.4A CN110516640B (en) 2019-08-30 2019-08-30 Vehicle re-identification method based on feature pyramid joint representation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910818186.4A CN110516640B (en) 2019-08-30 2019-08-30 Vehicle re-identification method based on feature pyramid joint representation

Publications (2)

Publication Number Publication Date
CN110516640A true CN110516640A (en) 2019-11-29
CN110516640B CN110516640B (en) 2022-09-30

Family

ID=68629852

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910818186.4A Active CN110516640B (en) 2019-08-30 2019-08-30 Vehicle re-identification method based on feature pyramid joint representation

Country Status (1)

Country Link
CN (1) CN110516640B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112418262A (en) * 2020-09-23 2021-02-26 上海市刑事科学技术研究院 Vehicle re-identification method, client and system
CN112598163A (en) * 2020-12-08 2021-04-02 国网河北省电力有限公司电力科学研究院 Grounding grid trenchless corrosion prediction model based on comparison learning and measurement learning

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107885764A (en) * 2017-09-21 2018-04-06 银江股份有限公司 Based on the quick Hash vehicle retrieval method of multitask deep learning
US20180114337A1 (en) * 2016-10-20 2018-04-26 Sun Yat-Sen University Method and system of detecting and recognizing a vehicle logo based on selective search
CN108108657A (en) * 2017-11-16 2018-06-01 浙江工业大学 A kind of amendment local sensitivity Hash vehicle retrieval method based on multitask deep learning

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180114337A1 (en) * 2016-10-20 2018-04-26 Sun Yat-Sen University Method and system of detecting and recognizing a vehicle logo based on selective search
CN107885764A (en) * 2017-09-21 2018-04-06 银江股份有限公司 Based on the quick Hash vehicle retrieval method of multitask deep learning
CN108108657A (en) * 2017-11-16 2018-06-01 浙江工业大学 A kind of amendment local sensitivity Hash vehicle retrieval method based on multitask deep learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
袁公萍等: "基于深度卷积神经网络的车型识别方法", 《浙江大学学报(工学版)》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112418262A (en) * 2020-09-23 2021-02-26 上海市刑事科学技术研究院 Vehicle re-identification method, client and system
CN112598163A (en) * 2020-12-08 2021-04-02 国网河北省电力有限公司电力科学研究院 Grounding grid trenchless corrosion prediction model based on comparison learning and measurement learning

Also Published As

Publication number Publication date
CN110516640B (en) 2022-09-30

Similar Documents

Publication Publication Date Title
CN112884064B (en) Target detection and identification method based on neural network
Qu et al. A crack detection algorithm for concrete pavement based on attention mechanism and multi-features fusion
CN110738146B (en) Target re-recognition neural network and construction method and application thereof
CN110309770B (en) Vehicle re-identification method based on four-tuple loss metric learning
CN112651262A (en) Cross-modal pedestrian re-identification method based on self-adaptive pedestrian alignment
CN109492589A (en) The recognition of face working method and intelligent chip merged by binary features with joint stepped construction
CN115063832A (en) Global and local feature-based cross-modal pedestrian re-identification method for counterstudy
CN110516640A (en) It is a kind of to combine the vehicle indicated discrimination method again based on feature pyramid
CN116342894A (en) GIS infrared feature recognition system and method based on improved YOLOv5
CN116977937A (en) Pedestrian re-identification method and system
Wang et al. Balanced one-stage object detection by enhancing the effect of positive samples
Ma et al. ISOD: improved small object detection based on extended scale feature pyramid network
CN113361475A (en) Multi-spectral pedestrian detection method based on multi-stage feature fusion information multiplexing
CN118015247A (en) Lightweight YOLO target detection method based on bidirectional multi-scale feature enhancement
CN117333908A (en) Cross-modal pedestrian re-recognition method based on attitude feature alignment
CN116935438A (en) Pedestrian image re-recognition method based on autonomous evolution of model structure
CN110110598A (en) The pedestrian of a kind of view-based access control model feature and space-time restriction recognition methods and system again
CN116665036A (en) RGB-infrared multi-source image target detection method based on single-mode auxiliary supervision and YOLOv5
CN116168418A (en) Multi-mode target perception and re-identification method for image
CN117994822B (en) Cross-mode pedestrian re-identification method based on auxiliary mode enhancement and multi-scale feature fusion
CN108650517B (en) Based on the determination method of group's image coding multiple reference images of object
Wang et al. Attentional single-shot network with multi-scale feature fusion for object detection in aerial images
Yang et al. Instance-aware detailed action labeling in videos
Wei et al. Interactive Transformer for Small Object Detection
Wang et al. Dual efficient self-attention network for multi-target detection in aerial imagery

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20191129

Assignee: Xiamen yunzhixin Intelligent Technology Co.,Ltd.

Assignor: HUAQIAO University

Contract record no.: X2024990000310

Denomination of invention: A vehicle re identification method based on feature pyramid joint representation

Granted publication date: 20220930

License type: Common License

Record date: 20240627

EE01 Entry into force of recordation of patent licensing contract