CN112906557A - Multi-granularity characteristic aggregation target re-identification method and system under multiple visual angles - Google Patents

Multi-granularity characteristic aggregation target re-identification method and system under multiple visual angles Download PDF

Info

Publication number
CN112906557A
CN112906557A CN202110183597.8A CN202110183597A CN112906557A CN 112906557 A CN112906557 A CN 112906557A CN 202110183597 A CN202110183597 A CN 202110183597A CN 112906557 A CN112906557 A CN 112906557A
Authority
CN
China
Prior art keywords
target
granularity
feature
hypergraph
target object
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
CN202110183597.8A
Other languages
Chinese (zh)
Other versions
CN112906557B (en
Inventor
彭德光
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.)
Chongqing Megalight Technology Co ltd
Original Assignee
Chongqing Megalight Technology Co ltd
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 Chongqing Megalight Technology Co ltd filed Critical Chongqing Megalight Technology Co ltd
Priority to CN202110183597.8A priority Critical patent/CN112906557B/en
Publication of CN112906557A publication Critical patent/CN112906557A/en
Application granted granted Critical
Publication of CN112906557B publication Critical patent/CN112906557B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
    • 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
    • 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
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
    • G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
    • G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
    • 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)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Multimedia (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Human Computer Interaction (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a method and a system for re-identifying a multi-granularity feature aggregation target under multiple visual angles, which comprise the following steps: constructing a multi-view neural network, and acquiring target characteristics of a target object from multiple views through the multi-view neural network; constructing a multi-granularity hypergraph based on the target characteristics of each target object in a set time period; inputting a target graph to be queried, and acquiring a neighboring feature set of the target image to be queried from the multi-granularity hypergraph; carrying out similarity comparison on the adjacent feature set of the target image to be inquired and the adjacent feature set of each target object in the multi-granularity hypergraph to obtain a target object re-identification result; the invention can effectively improve the re-identification precision.

Description

Multi-granularity characteristic aggregation target re-identification method and system under multiple visual angles
Technical Field
The invention relates to the field, in particular to a method and a system for re-identifying a multi-granularity feature aggregation target under multiple visual angles.
Background
Pedestrian re-identification based on video sequences is widely discussed because rich time information can be used for solving visual ambiguity, and the current classical method for video pedestrian re-identification is to adopt a deep learning method to a projection high-dimensional feature space for a video sequence and then perform identity matching sorting by calculating the distance between samples, and mainly comprises the steps of adopting a recurrent neural network to aggregate frame-level time features to represent video pedestrian features and using an optical flow field to extract video frame dynamic time information to learn the time features. The prior art has the following disadvantages: 1. the video learning based on the recurrent neural network cannot learn the most discriminant features, and the training of the model to the long-segment video is complex and takes long time. 2. The method for extracting the time characteristics by means of the light field exploration stream structure is easy to generate optical flow estimation errors due to the fact that adjacent frames of a certain section of video clip are not aligned. In order to solve the above problems, the invention 1 provides a video pedestrian re-identification method based on multi-granularity feature aggregation under multiple viewing angles, which captures multi-granularity spatial information and time information of a video sequence at the same time, and retains and enhances diversity discrimination feature representations of different spatial granularities by adopting a simple and efficient hypergraph construction mode.
Disclosure of Invention
In view of the problems in the prior art, the invention provides a method and a system for re-identifying a multi-granularity feature aggregation target under multiple viewing angles, and mainly solves the problems that the training time consumption is long and the accuracy is low in the conventional method.
In order to achieve the above and other objects, the present invention adopts the following technical solutions.
A multi-granularity feature aggregation target re-identification method under multiple visual angles comprises the following steps:
constructing a multi-view neural network, and acquiring target characteristics of a target object from multiple views through the multi-view neural network;
constructing a multi-granularity hypergraph based on the target characteristics of each target object in a set time period;
inputting a target graph to be queried, and acquiring a neighboring feature set of the target image to be queried from the multi-granularity hypergraph;
and carrying out similarity comparison on the adjacent feature set of the target image to be inquired and the adjacent feature set of each target object in the multi-granularity hypergraph to obtain a target object re-identification result.
Optionally, the multi-view neural network includes a convolutional neural network and a classification output layer, and the image is subjected to feature extraction by the convolutional neural network and then input to the classification output layer to obtain target feature outputs of different views.
Optionally, the multi-view neural network is pre-trained by inputting a set containing pre-labeled images with different views into the multi-view neural network, constructing a loss function through cross entropy, and updating network parameters by adopting back propagation.
Optionally, the loss function is expressed as:
Figure BDA0002942141000000021
wherein, yiFor the label corresponding to the angle of view,
Figure BDA0002942141000000022
for categorizing the prediction results, N is the number of views.
Optionally, the target object comprises a pedestrian or a vehicle.
Optionally, the obtaining a set of proximity features of the target image to be queried from the multi-granularity hypergraph includes:
calculating Euclidean distances among target features in the multi-granularity hypergraph, and acquiring the first K target features with the closest feature distances corresponding to the target image to be inquired;
and acquiring an adjacent set of each target feature in the K target features, and selecting the adjacent sets containing the corresponding features of the target image to be inquired from the adjacent sets to form an adjacent feature set of the target image to be inquired.
Optionally, the similarity comparison is performed between the neighboring feature set of the target image to be queried and the neighboring feature set of each target object in the multi-granularity hypergraph, and a target object re-identification result is obtained, including:
and measuring the similarity among the adjacent feature sets through the Jaccard distance, and selecting a target object corresponding to the adjacent feature set with the similarity reaching a set threshold value as re-recognition output.
Optionally, the similarity calculation manner is expressed as:
Figure BDA0002942141000000023
wherein, Ii，IjRespectively representing two frames of images, R (I)iK) represents an image IiThe set of neighboring features of (a).
A multi-granularity feature aggregation target re-identification system under multiple views comprises:
the network construction module is used for constructing a multi-view neural network and acquiring target characteristics of a target object from multiple views through the multi-view neural network;
the hypergraph construction module is used for constructing a multi-granularity hypergraph based on the target characteristics of each target object in a set time period;
the feature set acquisition module is used for inputting a target graph to be inquired and acquiring a neighboring feature set of the target image to be inquired from the multi-granularity hypergraph;
and the identification module is used for comparing the similarity of the adjacent feature set of the target image to be inquired with the adjacent feature set of each target object in the multi-granularity hypergraph to obtain a target object re-identification result.
As described above, the method and system for re-identifying the multi-granularity feature aggregation target under multiple viewing angles of the present invention have the following advantages.
Visual angle information is increased, and the problems of shielding, visual angle difference and the like are solved; and enhancing the re-identification precision through the neighbor feature set.
Drawings
Fig. 1 is a flowchart of a method for re-identifying a multi-granularity feature aggregation target under multiple viewing angles in an embodiment of the present invention.
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention, and the components related to the present invention are only shown in the drawings rather than drawn according to the number, shape and size of the components in actual implementation, and the type, quantity and proportion of the components in actual implementation may be changed freely, and the layout of the components may be more complicated.
Referring to fig. 1, the present invention provides a method for re-identifying a multi-granularity feature aggregation target under multiple viewing angles, which includes steps S01-S04.
In step S01, a multi-view neural network is constructed, and target features of a target object from multiple views are acquired through the multi-view neural network:
in one embodiment, the target object may include a pedestrian, a vehicle, etc., a video image including the target object is captured in advance, and the video sequence is acquired as an input of the multi-view neural network.
In one embodiment, the multi-view neural network comprises a convolutional neural network and a classification output layer, and after the image is subjected to feature extraction through the convolutional neural network, the image is input into the classification output layer to obtain target feature outputs of different views.
In one embodiment, a set containing pre-labeled images of different visual angles is input into a multi-visual-angle neural network, a loss function is constructed through cross entropy, network parameters are updated through back propagation, and the multi-visual-angle neural network is pre-trained.
Specifically, a ternary classification output layer is added after the traditional CNN, and a label image x is utilizediAs input, it corresponds to the view label yiAs supervisory signals, for the prediction results
Figure BDA0002942141000000041
And (3) carrying out supervision training by adopting cross entropy, wherein a cross entropy loss function can be expressed as:
Figure BDA0002942141000000042
the update calculation of the loss function is done using a forward and backward algorithm.
Extracting video frame characteristics;
for a video sequence I containing a picture I ═ I1,I2,...,ITAnd performing feature extraction on each image by using the constructed multi-view neural network, wherein the feature extraction can be specifically expressed as:
Fi＝CNN(Ii),i＝1,...,T,
wherein, FiA three-dimensional tensor, which represents dimensions C × H × W, represents a channel size, and represents the height and width of the feature map, respectively.
In step S02, a multi-granularity hypergraph is constructed based on the target features of the target objects within a set time period:
and dividing the image features extracted in the step S01 into p e {1,2,4,8} horizontal blocks in a horizontal division mode, and carrying out average combination on the divided feature maps to construct a partial feature vector. For each granularity, the entire sequence generates NpT × p partial level features, respectively
Figure BDA0002942141000000043
A first granularity of a video sequence comprises a single global feature vector, and other granularities comprise partial feature vectors.
First of all using vi∈Vp,i∈{1,2,...,NpThe points represent the preparation candidate nodes needed for constructing the hypergraph, and a group of hyperedges E are defined for capturing time informationpTo model short-term to long-term correlations in the hypergraph. Specifically, for any one candidate node viIs selected at time TtMost similar K adjacent nodes in the inner
Figure BDA0002942141000000051
Using the super edge, these K +1 nodes are as followsThe following are shown in the following table:
Figure BDA0002942141000000052
updating characteristics of the hypergraph;
for a certain node v of the hypergraphiDefinition of
Figure BDA0002942141000000053
And representing all the super edges related to the point, wherein the point related to one super edge has strong relevance, so that the super edge is defined by adopting the aggregation operation as follows:
Figure BDA0002942141000000054
wherein the content of the first and second substances,
Figure BDA0002942141000000055
denotes vjNode characteristics at the layer. Calculating the similarity of the association relationship between the node features and the association features of the super edges
Figure BDA0002942141000000056
Figure BDA0002942141000000057
Wherein the content of the first and second substances,
Figure BDA0002942141000000058
representing the similarity between features. In addition, SoftMax normalized similarity weight is adopted, and super-side information is aggregated to obtain the similarity weight through respective calculation
Figure BDA0002942141000000059
Figure BDA00029421410000000510
Figure BDA00029421410000000511
After the aggregated super-edge information is obtained, the node characteristics can be associated through a full connection layer:
Figure BDA00029421410000000512
wherein WlRepresenting the weight matrix and sigma the excitation equation. Therefore, repeating the update mechanism more than L times can calculate a series of output node characteristics
Figure BDA00029421410000000513
Hypergraph feature aggregation based on an attention mechanism;
after obtaining the final updated node characteristics for each hypergraph, it is considered that in one hypergraph, different nodes have different importance. For example: the lower the importance of the shielded part or background is, the better the feature discrimination is. Thus, we design a discriminant computation based on the attention mechanism, with nodes attentive to each hypergraph
Figure BDA0002942141000000061
Figure BDA0002942141000000062
Wherein, WuA weight matrix is represented. The hypergraph features can therefore be computed as a weight aggregation of node features:
Figure BDA0002942141000000063
minimizing retention loss aggregation multi-granularity hypergraphs based on mutual information;
to optimize the framework, cross-entropy loss and triplet state loss are employed to co-supervise the training process:
Figure BDA0002942141000000064
Figure BDA0002942141000000065
wherein y isiRepresentation feature
Figure BDA0002942141000000066
The labels of (1), N and C respectively represent the size of the mini-batch and the class number of the training set,
Figure BDA0002942141000000067
respectively representing a query sample and a positive sample and a negative sample thereof when the partition granularity is p. After training the model based on the two loss terms, each hypergraph will output a distinct graph-level feature.
In order to obtain the characteristics of the fused multi-granularity hypergraph information, mutual information minimization loss is adopted, mutual information among different hypergraph characteristics is reduced, and further the uncertainty of final video representation is increased by combining all the characteristics. Thus, for hypergraph features of different granularity p, a mutual information minimization loss is defined:
Figure BDA0002942141000000068
kappa is used to measure the mutual information established by the characteristics of different hypergraphs. And finally combining the loss functions of all parts as the formula (13), and adopting a forward-backward algorithm to complete the updating calculation of the loss functions.
Lall＝Lxent+Ltri+LMI
In step S03, a target image to be queried is input, and the set of neighboring features of the target image to be queried is obtained from the multi-granularity hypergraph:
in one embodiment, the Euclidean distance between the target features in the multi-granularity hypergraph is calculated, and the first K target features with the closest feature distance corresponding to the target image to be inquired are obtained;
and acquiring an adjacent set of each target feature in the K target features, and selecting the adjacent sets containing the corresponding features of the target image to be inquired from the adjacent sets to form an adjacent feature set of the target image to be inquired.
Specifically, the euclidean distance d between the hypergraph features obtained by step S03 is calculatedm(F′i,F′j) Calculating a neighbor set N (probe, k) corresponding to k nearest distances of the query image probe, where the set includes both positive samples and negative samples, and is defined as:
Figure BDA0002942141000000071
wherein the content of the first and second substances,
Figure BDA0002942141000000072
respectively, samples with Euclidean distances of 1 st, 2 nd and k th from probe. At the same time, for each of the neighbor sets N
Figure BDA0002942141000000073
And also has its own neighbor set N ', which are adjacent to each other if probes are included in N', and which are not adjacent to each other otherwise. Therefore, k-mutual adjacent set R of the probe can be obtained, and all elements in R are target objects which are adjacent to the probe.
R(probe,k)＝{(ti∈N(probe,k)∩(p∈N(ti，k))} (16)
The set can be regarded as k-mutual adjacent features of the probe, and is more suitable for similarity measurement between pedestrians compared with the hypergraph feature.
In step S04, the similarity between the neighboring feature set of the target image to be queried and the neighboring feature set of each target object in the multi-granularity hypergraph is compared to obtain a target object re-identification result:
in an embodiment, similarity between the neighboring feature sets is measured through a Jaccard distance, and a target object corresponding to the neighboring feature set with the similarity reaching a set threshold is selected as a re-recognition output, specifically:
for describing any two images I in detail from the perspective of collectioni，IjThe difference between the nearest neighbor sets defines the Jaccard distance between the two neighbor sets
Figure BDA0002942141000000074
And measuring the similarity between the target objects according to the distance, and re-identifying the query target object.
The embodiment also provides a system for re-identifying a multi-granularity feature aggregation target under multiple views, which is used for executing the method for re-identifying the multi-granularity feature aggregation target under multiple views in the method embodiment. Since the technical principle of the system embodiment is similar to that of the method embodiment, repeated description of the same technical details is omitted.
In one embodiment, a multi-granularity feature aggregation target re-identification system under multiple views comprises: the system comprises a network construction module, a hypergraph construction module, a feature set acquisition module and an identification module, wherein the network construction module is used for assisting in executing the step S01 in the embodiment of the method; the hypergraph construction module is used to assist in performing step S02 in the foregoing method embodiment; the feature set acquisition module is used for assisting in executing step S03 in the foregoing method embodiment; the identification module is used to assist in performing step S04 in the aforementioned method embodiments.
In summary, the multi-granularity feature aggregation target re-identification method and system under the visual angle adopt the ternary visual angle classification to enable the pedestrian features to contain visual angle information in subsequent processing, and solve the problems of shielding, visual angle difference and the like; the hypergraph neural network structure can simultaneously extract the spatial characteristics and the time dependence of video frames, and the hypergraph diversity corresponding to different spatial granularities can be reserved and enhanced by minimizing loss by using mutual information; the method of coding the adjacent k-ones improves the re-identification precision of the pedestrian, and makes up the defect that the hypergraph learning excessively focuses on the local information. Therefore, the invention effectively overcomes various defects in the prior art and has high industrial utilization value.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.

Claims (9)

1. A multi-granularity feature aggregation target re-identification method under multiple visual angles is characterized by comprising the following steps:
constructing a multi-view neural network, and acquiring target characteristics of a target object from multiple views through the multi-view neural network;
constructing a multi-granularity hypergraph based on the target characteristics of each target object in a set time period;
inputting a target graph to be queried, and acquiring a neighboring feature set of the target image to be queried from the multi-granularity hypergraph;
and carrying out similarity comparison on the adjacent feature set of the target image to be inquired and the adjacent feature set of each target object in the multi-granularity hypergraph to obtain a target object re-identification result.
2. The method according to claim 1, wherein the multi-view neural network comprises a convolutional neural network and a classification output layer, and after feature extraction is performed on the image by the convolutional neural network, the image is input to the classification output layer to obtain target feature outputs of different views.
3. The method for re-identifying the multi-granularity feature aggregation target under the multi-view angle according to claim 2, wherein different pre-labeled view angle image sets are input into the multi-view angle neural network, a loss function is constructed through cross entropy, network parameters are updated through back propagation, and the multi-view angle neural network is pre-trained.
4. The method for multi-granularity feature aggregation target re-identification under multiple views according to claim 3, wherein the loss function is expressed as:
Figure FDA0002942140990000011
wherein, yiFor the label corresponding to the angle of view,
Figure FDA0002942140990000012
for categorizing the prediction results, N is the number of views.
5. The method according to claim 1, wherein the target object comprises a pedestrian or a vehicle.
6. The method for re-identifying the multi-granularity feature aggregation target under the multi-view angle according to claim 1, wherein the step of obtaining the neighboring feature set of the target image to be queried from the multi-granularity hypergraph comprises the following steps:
calculating Euclidean distances among target features in the multi-granularity hypergraph, and acquiring the first K target features with the closest feature distances corresponding to the target image to be inquired;
and acquiring an adjacent set of each target feature in the K target features, and selecting the adjacent sets containing the corresponding features of the target image to be inquired from the adjacent sets to form an adjacent feature set of the target image to be inquired.
7. The method according to claim 1, wherein the comparing the similarity between the neighboring feature set of the target image to be queried and the neighboring feature set of each target object in the multi-granularity hypergraph to obtain a target object re-recognition result comprises:
and measuring the similarity among the adjacent feature sets through the Jaccard distance, and selecting a target object corresponding to the adjacent feature set with the similarity reaching a set threshold value as re-recognition output.
8. The method for re-identifying the multi-granularity feature aggregation target under the multi-view angle according to claim 7, wherein the similarity calculation mode is expressed as:
Figure FDA0002942140990000021
wherein, Ii，IjRespectively representing two frames of images, R (I)iK) represents an image IiThe set of neighboring features of (a).
9. A multi-granularity feature aggregation target re-identification system under multiple views is characterized by comprising:
the network construction module is used for constructing a multi-view neural network and acquiring target characteristics of a target object from multiple views through the multi-view neural network;
the hypergraph construction module is used for constructing a multi-granularity hypergraph based on the target characteristics of each target object in a set time period;
the feature set acquisition module is used for inputting a target graph to be inquired and acquiring a neighboring feature set of the target image to be inquired from the multi-granularity hypergraph;
and the identification module is used for comparing the similarity of the adjacent feature set of the target image to be inquired with the adjacent feature set of each target object in the multi-granularity hypergraph to obtain a target object re-identification result.
CN202110183597.8A 2021-02-08 2021-02-08 Multi-granularity feature aggregation target re-identification method and system under multi-view angle Active CN112906557B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110183597.8A CN112906557B (en) 2021-02-08 2021-02-08 Multi-granularity feature aggregation target re-identification method and system under multi-view angle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110183597.8A CN112906557B (en) 2021-02-08 2021-02-08 Multi-granularity feature aggregation target re-identification method and system under multi-view angle

Publications (2)

Publication Number Publication Date
CN112906557A true CN112906557A (en) 2021-06-04
CN112906557B CN112906557B (en) 2023-07-14

Family

ID=76123514

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110183597.8A Active CN112906557B (en) 2021-02-08 2021-02-08 Multi-granularity feature aggregation target re-identification method and system under multi-view angle

Country Status (1)

Country Link
CN (1) CN112906557B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114299128A (en) * 2021-12-30 2022-04-08 咪咕视讯科技有限公司 Multi-view positioning detection method and device
CN114419349A (en) * 2022-03-30 2022-04-29 中国科学技术大学 Image matching method and device

Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7428998B2 (en) * 2003-11-13 2008-09-30 Metrologic Instruments, Inc. Automatic hand-supportable image-based bar code symbol reader having image-processing based bar code reading subsystem employing simple decode image processing operations applied in an outwardly-directed manner referenced from the center of a captured narrow-area digital image of an object bearing a 1D bar code symbol
US8024193B2 (en) * 2006-10-10 2011-09-20 Apple Inc. Methods and apparatus related to pruning for concatenative text-to-speech synthesis
CN102663374A (en) * 2012-04-28 2012-09-12 北京工业大学 Multi-class Bagging gait recognition method based on multi-characteristic attribute
CN103959308A (en) * 2011-08-31 2014-07-30 Metaio有限公司 Method of matching image features with reference features
CN104061907A (en) * 2014-07-16 2014-09-24 中南大学 Viewing-angle greatly-variable gait recognition method based on gait three-dimensional contour matching synthesis
CN104281572A (en) * 2013-07-01 2015-01-14 中国科学院计算技术研究所 Target matching method and system based on mutual information
CN106096532A (en) * 2016-06-03 2016-11-09 山东大学 A kind of based on tensor simultaneous discriminant analysis across visual angle gait recognition method
CN106780551A (en) * 2016-11-18 2017-05-31 湖南拓视觉信息技术有限公司 A kind of Three-Dimensional Moving Targets detection method and system
CN109543602A (en) * 2018-11-21 2019-03-29 太原理工大学 A kind of recognition methods again of the pedestrian based on multi-view image feature decomposition
CN110738146A (en) * 2019-09-27 2020-01-31 华中科技大学 target re-recognition neural network and construction method and application thereof
CN111814584A (en) * 2020-06-18 2020-10-23 北京交通大学 Vehicle weight identification method under multi-view-angle environment based on multi-center measurement loss
CN112132014A (en) * 2020-09-22 2020-12-25 德州学院 Target re-identification method and system based on non-supervised pyramid similarity learning

Patent Citations (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7428998B2 (en) * 2003-11-13 2008-09-30 Metrologic Instruments, Inc. Automatic hand-supportable image-based bar code symbol reader having image-processing based bar code reading subsystem employing simple decode image processing operations applied in an outwardly-directed manner referenced from the center of a captured narrow-area digital image of an object bearing a 1D bar code symbol
US8024193B2 (en) * 2006-10-10 2011-09-20 Apple Inc. Methods and apparatus related to pruning for concatenative text-to-speech synthesis
CN103959308A (en) * 2011-08-31 2014-07-30 Metaio有限公司 Method of matching image features with reference features
CN102663374A (en) * 2012-04-28 2012-09-12 北京工业大学 Multi-class Bagging gait recognition method based on multi-characteristic attribute
CN104281572A (en) * 2013-07-01 2015-01-14 中国科学院计算技术研究所 Target matching method and system based on mutual information
CN104061907A (en) * 2014-07-16 2014-09-24 中南大学 Viewing-angle greatly-variable gait recognition method based on gait three-dimensional contour matching synthesis
CN106096532A (en) * 2016-06-03 2016-11-09 山东大学 A kind of based on tensor simultaneous discriminant analysis across visual angle gait recognition method
CN106780551A (en) * 2016-11-18 2017-05-31 湖南拓视觉信息技术有限公司 A kind of Three-Dimensional Moving Targets detection method and system
CN109543602A (en) * 2018-11-21 2019-03-29 太原理工大学 A kind of recognition methods again of the pedestrian based on multi-view image feature decomposition
CN110738146A (en) * 2019-09-27 2020-01-31 华中科技大学 target re-recognition neural network and construction method and application thereof
CN111814584A (en) * 2020-06-18 2020-10-23 北京交通大学 Vehicle weight identification method under multi-view-angle environment based on multi-center measurement loss
CN112132014A (en) * 2020-09-22 2020-12-25 德州学院 Target re-identification method and system based on non-supervised pyramid similarity learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
景陈勇等: "基于混合式协同训练的人体动作识别算法研究", 《计算机科学 》, pages 275 - 278 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114299128A (en) * 2021-12-30 2022-04-08 咪咕视讯科技有限公司 Multi-view positioning detection method and device
CN114419349A (en) * 2022-03-30 2022-04-29 中国科学技术大学 Image matching method and device
CN114419349B (en) * 2022-03-30 2022-07-15 中国科学技术大学 Image matching method and device

Also Published As

Publication number Publication date
CN112906557B (en) 2023-07-14

Similar Documents

Publication Publication Date Title
CN109948425B (en) Pedestrian searching method and device for structure-aware self-attention and online instance aggregation matching
CN110414368B (en) Unsupervised pedestrian re-identification method based on knowledge distillation
CN106547880B (en) Multi-dimensional geographic scene identification method fusing geographic area knowledge
CN109344285B (en) Monitoring-oriented video map construction and mining method and equipment
CN105809672B (en) A kind of image multiple target collaboration dividing method constrained based on super-pixel and structuring
CN112507901B (en) Unsupervised pedestrian re-identification method based on pseudo tag self-correction
CN109165540B (en) Pedestrian searching method and device based on prior candidate box selection strategy
CN111259786A (en) Pedestrian re-identification method based on synchronous enhancement of appearance and motion information of video
CN109871875B (en) Building change detection method based on deep learning
CN105184260B (en) A kind of image characteristic extracting method and pedestrian detection method and device
CN109743642B (en) Video abstract generation method based on hierarchical recurrent neural network
CN113313123B (en) Glance path prediction method based on semantic inference
CN112633382A (en) Mutual-neighbor-based few-sample image classification method and system
CN114399644A (en) Target detection method and device based on small sample
CN109598220A (en) A kind of demographic method based on the polynary multiple dimensioned convolution of input
CN109919112B (en) Method for detecting distribution and counting of flowing crowds in complex scene
CN112906557A (en) Multi-granularity characteristic aggregation target re-identification method and system under multiple visual angles
CN112270286A (en) Shadow interference resistant monochrome video target tracking method
CN115690549A (en) Target detection method for realizing multi-dimensional feature fusion based on parallel interaction architecture model
CN104463962B (en) Three-dimensional scene reconstruction method based on GPS information video
CN105631858B (en) Image object method of counting based on sample block
Abdullah et al. Vehicle counting using deep learning models: a comparative study
CN112329662B (en) Multi-view saliency estimation method based on unsupervised learning
CN111611919B (en) Road scene layout analysis method based on structured learning
CN109740405B (en) Method for detecting front window difference information of non-aligned similar vehicles

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

Address after: 400000 6-1, 6-2, 6-3, 6-4, building 7, No. 50, Shuangxing Avenue, Biquan street, Bishan District, Chongqing

Applicant after: CHONGQING ZHAOGUANG TECHNOLOGY CO.,LTD.

Address before: 400000 2-2-1, 109 Fengtian Avenue, tianxingqiao, Shapingba District, Chongqing

Applicant before: CHONGQING ZHAOGUANG TECHNOLOGY CO.,LTD.

CB02 Change of applicant information
GR01 Patent grant
GR01 Patent grant