CN116258751A - Security patrol management method, system and equipment based on unsupervised learning - Google Patents

Security patrol management method, system and equipment based on unsupervised learning Download PDF

Info

Publication number
CN116258751A
CN116258751A CN202310506915.9A CN202310506915A CN116258751A CN 116258751 A CN116258751 A CN 116258751A CN 202310506915 A CN202310506915 A CN 202310506915A CN 116258751 A CN116258751 A CN 116258751A
Authority
CN
China
Prior art keywords
security personnel
pedestrian
feature
patrol
matching
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
Application number
CN202310506915.9A
Other languages
Chinese (zh)
Inventor
郭金林
赵研承
谢毓湘
康来
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.)
National University of Defense Technology
Original Assignee
National University of Defense Technology
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 National University of Defense Technology filed Critical National University of Defense Technology
Priority to CN202310506915.9A priority Critical patent/CN116258751A/en
Publication of CN116258751A publication Critical patent/CN116258751A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
    • 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
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • 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
    • G06V10/62—Extraction of image or video features relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G06V10/763—Non-hierarchical techniques, e.g. based on statistics of modelling distributions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G06V10/7753—Incorporation of unlabelled data, e.g. multiple instance learning [MIL]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • 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
    • G07—CHECKING-DEVICES
    • G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
    • G07C1/00—Registering, indicating or recording the time of events or elapsed time, e.g. time-recorders for work people
    • G07C1/20—Checking timed patrols, e.g. of watchman
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30241—Trajectory
    • 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)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Multimedia (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Human Computer Interaction (AREA)
  • Probability & Statistics with Applications (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Alarm Systems (AREA)

Abstract

The application relates to a security patrol management method, system and equipment based on unsupervised learning. The method comprises the following steps: training the feature extraction network through the pedestrian features extracted by the feature extraction network and pseudo tags obtained by clustering the pedestrian features, and realizing comparison and identification of security personnel in an unsupervised form according to the trained feature extraction network; the security personnel identification result is divided into a high-resolution detection area and a low-resolution detection area, and two-stage matching tracking is performed according to the high-resolution detection area and the low-resolution detection area, so that the security personnel motion trail matching result is obtained. By means of the method, whether the motion trail matching result of the security personnel is the same as the specified patrol trail of the security personnel can be judged, the patrol state of the security personnel can be judged, and when the patrol of the security personnel is wrong or not in place, a prompt signal is given, so that effective management of the patrol state of the security personnel under the condition of low cost is achieved.

Description

Security patrol management method, system and equipment based on unsupervised learning
Technical Field
The application relates to the technical field of security personnel patrol management, in particular to a security patrol management method, system and equipment based on unsupervised learning.
Background
Patrol is an important job of security, and is an important method for preventing security accidents. In recent years, with the improvement of economic level, the security industry enters a rapid development stage, and a plurality of problems also occur in the rapid development process. Security patrols typically have pre-fixed routes and areas. Because patrol work is simpler, individual security staff has the problems of low execution power, low work responsibility and the like, the importance of patrol is not realized, and the security accidents are often caused by lazy patrol or patrol according to a specified route. In the management process of security companies, the management system is not performed in place, and an effective supervision and management tool is usually lacking.
The existing security personnel patrol management system generally adopts a pedestrian re-identification algorithm to conduct positioning identification on security personnel based on a computer vision solution, so that security personnel patrol state supervision is achieved, but the algorithm is trained on a labeled data set, a large amount of high-quality data which are manually marked are required for training the algorithm, high cost is required for obtaining the marked data in actual production and application, and marking data obtaining difficulty is high.
Disclosure of Invention
Based on the above, it is necessary to provide a security patrol management method, system and device based on unsupervised learning, which can effectively manage the patrol state of security personnel without a large amount of manual labeling data.
The security patrol management system based on the unsupervised learning is used for judging the patrol state of the security personnel and realizing the patrol management of the security personnel by comparing the motion trail matching result of the security personnel with the specified patrol trail of the security personnel or not, and comprises: the system comprises a pedestrian detection module, a network training module, a security personnel identification module and a motion trail matching module;
the method comprises the following steps:
the pedestrian detection results at different moments in the patrol area are obtained according to the pedestrian detection module, and the pedestrian detection results are input into a pre-constructed feature extraction network to be processed, so that pedestrian features are obtained; wherein the pedestrian features include global features and local features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features, and the second branch is used for extracting local features;
in a network training module, clustering pedestrian features by adopting a density clustering algorithm, distributing a corresponding pseudo tag for each pedestrian according to a clustering result, and training a feature extraction network according to an optimization function constructed by the pedestrian features and the pseudo tags to obtain a trained feature extraction network;
inputting the pedestrian detection result to be identified into a trained feature extraction network for feature extraction according to the security personnel identification module to obtain the final feature of the pedestrian to be identified, and selecting the security personnel feature with the maximum similarity as the security personnel identification result by calculating and comparing the similarity of the final feature of the pedestrian to be identified and the security personnel feature in the database;
in the motion trail matching module, security personnel identification results at different moments are divided into a high-score detection area and a low-score detection area according to a preset similarity threshold, and the current moment positions of security personnel predicted by a security personnel patrol trail set are respectively matched and associated with the high-score detection area and the low-score detection area to obtain security personnel motion trail matching results.
In one embodiment, the pedestrian detection module obtains pedestrian detection results at different moments in a patrol area, and inputs the pedestrian detection results into a pre-constructed feature extraction network for processing to obtain pedestrian features, including:
detecting pedestrians in the patrol area by using a Yolov7 pedestrian detector according to the pedestrian detection module, and obtaining pedestrian detection results at different moments;
inputting the pedestrian detection result into a pre-constructed feature extraction network for processing to obtain pedestrian features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features of pedestrian detection results
Figure SMS_1
The second branch is used for dividing the pedestrian detection result into three equal parts according to the ordinate direction and respectively extracting first local features of the three equal parts of pedestrian detection result +.>
Figure SMS_2
Second local feature->
Figure SMS_3
And third local feature->
Figure SMS_4
。
In one embodiment, in the network training module, a density clustering algorithm is used to cluster pedestrian features, and a corresponding pseudo tag is allocated to each pedestrian according to a clustering result, including:
in the network training module, a DbScan density clustering algorithm is adopted for global features
Figure SMS_5
First local feature->
Figure SMS_6
Second local feature->
Figure SMS_7
Third local feature->
Figure SMS_8
Clustering, increasing the number of clustering centers by adopting a method of manually checking the clustering result, storing the clustering centers as final clustering results, and distributing a corresponding pseudo tag to each pedestrian according to the final clustering results, wherein the pseudo tag is expressed as
Figure SMS_9
；
wherein ,
Figure SMS_10
respectively represent +.>
Figure SMS_11
And the individual person obtains a cluster label according to the global feature, the first local feature, the second local feature and the third local feature.
In one embodiment, training the feature extraction network according to an optimization function constructed by pedestrian features and pseudo tags to obtain a trained feature extraction network, including:
training the feature extraction network according to the pedestrian features and the optimization function constructed by the pseudo tag by taking the pseudo tag as a data tag, wherein the feature extraction network used in each training is the feature extraction network obtained in the previous training until convergence to obtain a trained feature extraction network; wherein the optimization function is expressed as
Figure SMS_12
；
wherein ,
Figure SMS_13
representing a ternary loss function, +.>
Figure SMS_14
、
Figure SMS_15
、
Figure SMS_16
and
Figure SMS_17
The cluster labels obtained according to the global feature, the first local feature, the second local feature and the third local feature are respectively represented.
In one embodiment, according to the security personnel identification module, inputting a pedestrian detection result to be identified into a trained feature extraction network for feature extraction to obtain a final feature of the pedestrian to be identified, and selecting the security personnel feature with the largest similarity as the security personnel identification result by calculating and comparing the similarity of the final feature of the pedestrian to be identified and the security personnel feature in the database, wherein the method comprises the following steps:
inputting a pedestrian detection result to be identified into a trained feature extraction network according to a security personnel identification module to perform feature extraction to obtain a global feature of the pedestrian to be identified and a local feature to be identified, and splicing the global feature of the pedestrian to be identified and the local feature to be identified to obtain a final feature of the pedestrian to be identified;
the cosine distance between the final characteristics of the pedestrians to be identified and the security personnel characteristics in the database is calculated and compared, and the nearest security personnel characteristics with the cosine distance are selected as security personnel identification results; the cosine distance nearest represents the maximum similarity between the final characteristics of the pedestrians to be identified and the characteristics of security personnel in the database.
In one embodiment, in the motion trail matching module, security personnel identification results at different moments are divided into a high-score detection area and a low-score detection area according to a preset similarity threshold, and the current moment positions of security personnel predicted by a security personnel patrol trail set are respectively matched and associated with the high-score detection area and the low-score detection area to obtain a security personnel motion trail matching result, wherein the security personnel motion trail matching result is expressed as:
in the motion trail matching module, according to a preset similarity threshold value
Figure SMS_18
Dividing security personnel identification results at different moments into high-score detection areas ++>
Figure SMS_19
And low-resolution detection region->
Figure SMS_20
Wherein, is higher than threshold +.>
Figure SMS_21
The security personnel identification result of (2) is classified as +.>
Figure SMS_22
Below threshold +.>
Figure SMS_23
The identification result of security personnel is classified as +.>
Figure SMS_24
；
The security personnel patrol track set T uses the current time position of the security personnel predicted by the Kalman tracker and Gao Fenjian area
Figure SMS_25
Performing first matching association to obtain a first-stage motion trail matching result; wherein, the high-resolution detection area which is not successfully matched is +.>
Figure SMS_26
Stored in the remaining track->
Figure SMS_27
In (a) and (b);
the security personnel patrol track set T is respectively connected with the low-resolution detection area by the current time position of the security personnel predicted by the Kalman tracker
Figure SMS_28
And remaining track->
Figure SMS_29
Performing second matching association to obtain a second-stage motion trail matching result; the method for matching association of the two times is consistent, and security personnel patrol tracks which are not successfully matched in the two times are stored in a mismatch track set +.>
Figure SMS_30
In the method, the detection area which is not successfully matched twice is deleted;
and combining the first-stage motion trail matching result and the second-stage motion trail matching result to obtain the security personnel motion trail matching result.
In one embodiment, the security personnel patrol track set T predicts the current time position of the security personnel and Gao Fenjian area by using a Kalman tracker
Figure SMS_31
Performing first matching association to obtain a first-stage motion trail matching result, wherein the first-stage motion trail matching result comprises the following steps of:
calculating the current time position and Gao Fenjian area of security personnel predicted by the security personnel patrol track set T through a Kalman tracker
Figure SMS_32
Setting a cross-over ratio threshold, rejecting matching for the region cross-over ratio smaller than the cross-over ratio threshold, and matching for the region cross-over ratio not smaller than the cross-over ratio threshold to obtain a matching matrix;
and carrying out matching solution on the matching matrix according to the Hungary algorithm to obtain a first-stage motion trail matching result.
In one embodiment, when matching is performed, for a set of non-matching tracks
Figure SMS_33
If there is a mismatch trace in (a) and if there is timeIf the length of the existing time exceeds the specified length, deleting the unmatched track; otherwise, continuously storing the unmatched tracks, and collecting the unmatched tracks from the unmatched track set when the unmatched tracks are successfully matched in the subsequent matching association process
Figure SMS_34
Transferring to a security personnel patrol track set T; wherein the prescribed duration of the presence time is 2 seconds.
A security patrol management system based on unsupervised learning, which judges patrol states of security personnel by comparing motion trail matching results of the security personnel with prescribed patrol trails of the security personnel to realize patrol management of the security personnel, the system comprising:
the pedestrian detection module is used for acquiring pedestrian detection results at different moments in the patrol area, inputting the pedestrian detection results into a pre-constructed feature extraction network for processing, and obtaining pedestrian features; wherein the pedestrian features include global features and local features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features, and the second branch is used for extracting local features;
the network training module is used for clustering pedestrian features by adopting a density clustering algorithm, distributing a corresponding pseudo tag for each pedestrian according to a clustering result, and training the feature extraction network according to an optimization function constructed by the pedestrian features and the pseudo tags to obtain a trained feature extraction network;
the security personnel identification module is used for inputting the pedestrian detection result to be identified into the trained feature extraction network to perform feature extraction to obtain the final feature of the pedestrian to be identified, calculating and comparing the similarity between the final feature of the pedestrian to be identified and the security personnel feature in the database, and selecting the security personnel feature with the maximum similarity as the security personnel identification result;
the motion trail matching module is used for dividing security personnel identification results at different moments into a high-score detection area and a low-score detection area according to a preset similarity threshold value, and respectively matching and associating the current moment positions of security personnel predicted by the security personnel patrol trail set with the high-score detection area and the low-score detection area to obtain a security personnel motion trail matching result.
A computer device comprising a memory storing a computer program and a processor which when executing the computer program performs the steps of:
the pedestrian detection results at different moments in the patrol area are obtained according to the pedestrian detection module, and the pedestrian detection results are input into a pre-constructed feature extraction network to be processed, so that pedestrian features are obtained; wherein the pedestrian features include global features and local features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features, and the second branch is used for extracting local features;
in a network training module, clustering pedestrian features by adopting a density clustering algorithm, distributing a corresponding pseudo tag for each pedestrian according to a clustering result, and training a feature extraction network according to an optimization function constructed by the pedestrian features and the pseudo tags to obtain a trained feature extraction network;
inputting the pedestrian detection result to be identified into a trained feature extraction network for feature extraction according to the security personnel identification module to obtain the final feature of the pedestrian to be identified, and selecting the security personnel feature with the maximum similarity as the security personnel identification result by calculating and comparing the similarity of the final feature of the pedestrian to be identified and the security personnel feature in the database;
in the motion trail matching module, security personnel identification results at different moments are divided into a high-score detection area and a low-score detection area according to a preset similarity threshold, and the current moment positions of security personnel predicted by a security personnel patrol trail set are respectively matched and associated with the high-score detection area and the low-score detection area to obtain security personnel motion trail matching results.
According to the security patrol management method, the security patrol management system and the security patrol management equipment based on the unsupervised learning, the feature extraction network is trained through the pedestrian features extracted by the feature extraction network and the pseudo tags obtained by clustering the pedestrian features, so that comparison and identification of security personnel in an unsupervised form are realized according to the trained feature extraction network, and the cost of detection and identification of the security personnel is reduced; and the security personnel identification result is subjected to two-stage matching tracking, so that the security personnel motion trail matching result is obtained, and the security personnel motion trail tracking accuracy is improved. After the security patrol is finished, the security patrol management system can judge the patrol state of the security personnel by comparing the motion trail matching result of the security personnel with the specified patrol trail of the security personnel, and give out a prompt signal when the security personnel patrol is wrong or not in place, so that the effective management of the patrol state of the security personnel under the condition of low cost is realized.
Drawings
FIG. 1 is a flow diagram of a security patrol management method based on unsupervised learning in one embodiment;
FIG. 2 is a block diagram of a security patrol management system based on unsupervised learning in one embodiment;
FIG. 3 is an internal block diagram of a computer device in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application will be further described in detail with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the present application.
In one embodiment, as shown in fig. 1, there is provided a security patrol management method based on unsupervised learning, which is applied to a security patrol management system based on unsupervised learning, and the security patrol management system based on unsupervised learning judges patrol states of security personnel by comparing motion trail matching results of the security personnel with prescribed patrol trails of the security personnel, so as to implement patrol management of the security personnel, and the method includes the following steps:
102, acquiring pedestrian detection results at different moments in a patrol area according to a pedestrian detection module, and inputting the pedestrian detection results into a pre-constructed feature extraction network for processing to obtain pedestrian features; wherein the pedestrian features include global features and local features; the feature extraction network includes a first branch for extracting global features and a second branch for extracting local features.
The feature extraction network adopts a multi-task framework, a main network of the feature extraction network adopts ResNet11 (residual error network), two branches are designed on the basis of the residual error network, wherein the first branch is output by the ResNet11, 512-dimensional global features are generated through convolution pooling operation, and the second branch also adopts a convolution pooling mode to respectively generate three 512-dimensional local features which respectively represent different parts of the body of a pedestrian.
And 104, clustering pedestrian features by adopting a density clustering algorithm in a network training module, distributing a corresponding pseudo tag for each pedestrian according to a clustering result, and training a feature extraction network according to an optimization function constructed by the pedestrian features and the pseudo tags to obtain a trained feature extraction network.
Specifically, the present application clusters pedestrian features using a DbScan (Density-Based Spatial Clustering of Applications with Noise, noise based spatial clustering) algorithm that defines clusters as the largest set of points that are connected in Density, is able to divide areas with a sufficiently high Density into clusters, and can find clusters of arbitrary shape in a spatial database of noise.
And 106, inputting the pedestrian detection result to be identified into a trained feature extraction network to perform feature extraction according to the security personnel identification module, obtaining the final feature of the pedestrian to be identified, and selecting the security personnel feature with the maximum similarity as the security personnel identification result by calculating and comparing the similarity of the final feature of the pedestrian to be identified and the security personnel feature in the database.
It can be understood that the pseudo tag is used as a data mark during training, and the feature extraction network is trained according to the optimization function until convergence, so as to obtain a trained feature extraction network, wherein the feature extraction network used during each training is the feature extraction network obtained during the previous training, and the feature extraction network with pre-training weight is adopted during the first training to extract the pedestrian feature.
And step 108, in the motion trail matching module, the security personnel identification results at different moments are divided into a high-score detection area and a low-score detection area according to a preset similarity threshold, and the current moment positions of the security personnel predicted by the security personnel patrol trail set are respectively matched and associated with the high-score detection area and the low-score detection area to obtain the security personnel motion trail matching result.
It can be understood that in order to obtain the motion trail of each security personnel, the motion trail of the security personnel is tracked based on the security personnel identification result obtained by the identification, and the tracking accuracy is improved by adopting a two-stage matching tracking algorithm.
In one embodiment, the pedestrian detection module obtains pedestrian detection results at different moments in a patrol area, and inputs the pedestrian detection results into a pre-constructed feature extraction network for processing to obtain pedestrian features, including:
detecting pedestrians in the patrol area by using a Yolov7 pedestrian detector according to the pedestrian detection module, and obtaining pedestrian detection results at different moments;
inputting the pedestrian detection result into a pre-constructed feature extraction network for processing to obtain pedestrian features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features of pedestrian detection results
Figure SMS_35
The second branch is used for dividing the pedestrian detection result into three equal parts according to the ordinate direction and respectively extracting first local features of the three equal parts of pedestrian detection result +.>
Figure SMS_36
Second local feature->
Figure SMS_37
And third local feature->
Figure SMS_38
Respectively correspond to the upper, middle and lower parts of the pedestrian.
In one embodiment, in the network training module, a density clustering algorithm is used to cluster pedestrian features, and a corresponding pseudo tag is allocated to each pedestrian according to a clustering result, including:
in the network training module, a DbScan density clustering algorithm is adopted for global features
Figure SMS_39
First local feature->
Figure SMS_40
Second local feature->
Figure SMS_41
Third local feature->
Figure SMS_42
Clustering, increasing the number of clustering centers by adopting a method of manually checking the clustering result, storing the clustering centers as final clustering results, and distributing a corresponding pseudo tag to each pedestrian according to the final clustering results, wherein the pseudo tag is expressed as
Figure SMS_43
；
wherein ,
Figure SMS_44
respectively represent +.>
Figure SMS_45
And the individual person obtains a cluster label according to the global feature, the first local feature, the second local feature and the third local feature.
In one embodiment, training the feature extraction network according to an optimization function constructed by pedestrian features and pseudo tags to obtain a trained feature extraction network, including:
training the feature extraction network according to the pedestrian features and the optimization function constructed by the pseudo tag by taking the pseudo tag as a data tag, wherein the feature extraction network used in each training is the feature extraction network obtained in the previous training until convergence to obtain a trained feature extraction network; wherein the optimization function is expressed as
Figure SMS_46
；
wherein ,
Figure SMS_47
representing a ternary loss function, randomly selecting categories with the same and different pseudo tags for a batch of data to construct a triplet in a specific calculation process, and completing calculation of loss +.>
Figure SMS_48
、
Figure SMS_49
、
Figure SMS_50
and
Figure SMS_51
The cluster labels obtained according to the global feature, the first local feature, the second local feature and the third local feature are respectively represented.
In one embodiment, according to the security personnel identification module, inputting a pedestrian detection result to be identified into a trained feature extraction network for feature extraction to obtain a final feature of the pedestrian to be identified, and selecting the security personnel feature with the largest similarity as the security personnel identification result by calculating and comparing the similarity of the final feature of the pedestrian to be identified and the security personnel feature in the database, wherein the method comprises the following steps:
inputting a pedestrian detection result to be identified into a trained feature extraction network according to a security personnel identification module to perform feature extraction to obtain a global feature of the pedestrian to be identified and a local feature to be identified, and splicing the global feature of the pedestrian to be identified and the local feature to be identified to obtain a 2048-dimensional final feature of the pedestrian to be identified as a final representation of the pedestrian;
the cosine distance between the final characteristics of the pedestrians to be identified and the security personnel characteristics in the database is calculated and compared, and the nearest security personnel characteristics with the cosine distance are selected as security personnel identification results; the cosine distance nearest represents the maximum similarity between the final characteristics of the pedestrians to be identified and the characteristics of security personnel in the database.
In one embodiment, in the motion trail matching module, security personnel identification results at different moments are divided into a high-score detection area and a low-score detection area according to a preset similarity threshold, and the current moment positions of security personnel predicted by a security personnel patrol trail set are respectively matched and associated with the high-score detection area and the low-score detection area to obtain a security personnel motion trail matching result, wherein the security personnel motion trail matching result is expressed as:
in the motion trail matching module, according to a preset similarity threshold value
Figure SMS_52
Dividing security personnel identification results at different moments into high-score detection areas ++>
Figure SMS_53
And low-resolution detection region->
Figure SMS_54
Wherein, is higher than threshold +.>
Figure SMS_55
The security personnel identification result of (2) is classified as +.>
Figure SMS_56
Below threshold +.>
Figure SMS_57
The identification result of security personnel is classified as +.>
Figure SMS_58
；
The security personnel patrol track set T is used for predicting the current time position and the current time position of the security personnel by using a Kalman trackerHigh-resolution detection region
Figure SMS_59
Performing first matching association to obtain a first-stage motion trail matching result; wherein, the high-resolution detection area which is not successfully matched is +.>
Figure SMS_60
Stored in the remaining track->
Figure SMS_61
In (a) and (b);
the security personnel patrol track set T is respectively connected with the low-resolution detection area by the current time position of the security personnel predicted by the Kalman tracker
Figure SMS_62
And remaining track->
Figure SMS_63
Performing second matching association to obtain a second-stage motion trail matching result; the method for matching association of the two times is consistent, and security personnel patrol tracks which are not successfully matched in the two times are stored in a mismatch track set +.>
Figure SMS_64
In the method, the detection area which is not successfully matched twice is deleted;
and combining the first-stage motion trail matching result and the second-stage motion trail matching result to obtain the security personnel motion trail matching result.
Specifically, before the motion trail matching, an initialization process is further performed, including: collecting video frames from a camera; initializing a Kalman tracker (KF); setting a threshold value
Figure SMS_65
To represent that the security personnel identified similarity initializes the security personnel patrol track set T to be an empty set. After the motion trail matching is carried out, the Kalman tracker is updated according to the motion trail matching result of security personnel.
In one of themIn an embodiment, the security personnel patrol track set T predicts the current time position of the security personnel and Gao Fenjian area by using a Kalman tracker
Figure SMS_66
Performing first matching association to obtain a first-stage motion trail matching result, wherein the first-stage motion trail matching result comprises the following steps of:
calculating the current time position and Gao Fenjian area of security personnel predicted by the security personnel patrol track set T through a Kalman tracker
Figure SMS_67
Setting an intersection ratio threshold value, refusing to match if the intersection ratio of the region is smaller than the intersection ratio threshold value, and matching if the intersection ratio of the region is not smaller than the intersection ratio threshold value, so as to obtain a matching matrix; specifically, an IOU threshold of 0.2 is set, and a match is rejected for IOUs less than 0.2.
And carrying out matching solution on the matching matrix according to the Hungary algorithm to obtain a first-stage motion trail matching result.
It will be understood that the intersection ratio means the ratio of the intersection and the union, and is used to evaluate the area overlap ratio of two geometric figures, and the larger the intersection ratio, the more relevant the two geometric figures.
In one embodiment, when matching is performed, for a set of non-matching tracks
Figure SMS_68
If the existing time exceeds the specified existing time length, deleting the unmatched track; otherwise, continuously storing the unmatched tracks, and collecting the unmatched tracks from the unmatched track set when the unmatched tracks are successfully matched in the subsequent matching association process
Figure SMS_69
Transferring to a security personnel patrol track set T; wherein the prescribed duration of the presence time is 2 seconds.
It should be understood that, although the steps in the flowchart of fig. 1 are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 1 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, nor do the order in which the sub-steps or stages are performed necessarily performed in sequence, but may be performed alternately or alternately with at least a portion of other steps or sub-steps of other steps.
In one embodiment, as shown in fig. 2, there is provided an unsupervised learning-based security patrol management system for determining a patrol status of a security personnel by comparing a result of matching a motion trail of the security personnel with a specified patrol trail of the security personnel, thereby implementing patrol management of the security personnel, the system comprising: the system comprises a pedestrian detection module 201, a network training module 202, a security personnel identification module 203 and a motion trail matching module 204;
the pedestrian detection module 201 is configured to obtain pedestrian detection results at different moments in the patrol area, and input the pedestrian detection results into a pre-constructed feature extraction network for processing, so as to obtain pedestrian features; wherein the pedestrian features include global features and local features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features, and the second branch is used for extracting local features;
the network training module 202 is configured to cluster pedestrian features by using a density clustering algorithm, allocate a corresponding pseudo tag to each pedestrian according to a clustering result, and train the feature extraction network according to an optimization function constructed by the pedestrian features and the pseudo tags to obtain a trained feature extraction network;
the security personnel identification module 203 is configured to input a pedestrian detection result to be identified into a trained feature extraction network to perform feature extraction, obtain a final feature of the pedestrian to be identified, calculate and compare similarity between the final feature of the pedestrian to be identified and security personnel features in the database, and select the security personnel feature with the largest similarity as a security personnel identification result;
the motion trail matching module 204 is configured to divide security personnel identification results at different moments into a high-score detection area and a low-score detection area according to a preset similarity threshold, and match and correlate the current moment positions of security personnel predicted by the security personnel patrol trail set with the high-score detection area and the low-score detection area respectively, so as to obtain a security personnel motion trail matching result.
Specific limitations regarding the security patrol management system based on the unsupervised learning can be found in the above description of the security patrol management method based on the unsupervised learning, and will not be described herein. The above-mentioned various modules in the security patrol management system based on the unsupervised learning may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a terminal, and the internal structure of which may be as shown in fig. 3. The computer device includes a processor, a memory, a network interface, a display screen, and an input system connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by the processor, implements a security patrol management method based on unsupervised learning. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input system of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.
It will be appreciated by those skilled in the art that the structure shown in fig. 3 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided comprising a memory storing a computer program and a processor that when executing the computer program performs the steps of:
the pedestrian detection results at different moments in the patrol area are obtained according to the pedestrian detection module, and the pedestrian detection results are input into a pre-constructed feature extraction network to be processed, so that pedestrian features are obtained; wherein the pedestrian features include global features and local features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features, and the second branch is used for extracting local features;
in a network training module, clustering pedestrian features by adopting a density clustering algorithm, distributing a corresponding pseudo tag for each pedestrian according to a clustering result, and training a feature extraction network according to an optimization function constructed by the pedestrian features and the pseudo tags to obtain a trained feature extraction network;
inputting the pedestrian detection result to be identified into a trained feature extraction network for feature extraction according to the security personnel identification module to obtain the final feature of the pedestrian to be identified, and selecting the security personnel feature with the maximum similarity as the security personnel identification result by calculating and comparing the similarity of the final feature of the pedestrian to be identified and the security personnel feature in the database;
in the motion trail matching module, security personnel identification results at different moments are divided into a high-score detection area and a low-score detection area according to a preset similarity threshold, and the current moment positions of security personnel predicted by a security personnel patrol trail set are respectively matched and associated with the high-score detection area and the low-score detection area to obtain security personnel motion trail matching results.
The technical features of the above embodiments may be arbitrarily combined, and all possible combinations of the technical features in the above embodiments are not described for brevity of description, however, as long as there is no contradiction between the combinations of the technical features, they should be considered as the scope of the description.
The above examples only represent a few embodiments of the present application, which are described in more detail and are not to be construed as limiting the scope of the present application. It should be noted that it would be apparent to those skilled in the art that various modifications and improvements could be made without departing from the spirit of the present application, which would be within the scope of the present application. Accordingly, the scope of protection of the present application shall be subject to the appended claims.

Claims (10)

1. The security patrol management method based on the unsupervised learning is characterized by being applied to a security patrol management system based on the unsupervised learning, wherein the security patrol management system based on the unsupervised learning judges the patrol state of security personnel by comparing the motion trail matching result of the security personnel with the specified patrol trail of the security personnel to realize the patrol management of the security personnel, and the security patrol management system based on the unsupervised learning comprises the following steps: the system comprises a pedestrian detection module, a network training module, a security personnel identification module and a motion trail matching module;
the method comprises the following steps:
the pedestrian detection module acquires pedestrian detection results at different moments in a patrol area, and inputs the pedestrian detection results into a pre-constructed feature extraction network for processing to obtain pedestrian features; wherein the pedestrian features include global features and local features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features, and the second branch is used for extracting local features;
in the network training module, clustering the pedestrian characteristics by adopting a density clustering algorithm, distributing a corresponding pseudo tag for each pedestrian according to a clustering result, and training the characteristic extraction network according to an optimization function constructed by the pedestrian characteristics and the pseudo tags to obtain a trained characteristic extraction network;
inputting the pedestrian detection result to be identified into a trained feature extraction network to perform feature extraction according to the security personnel identification module to obtain the final feature of the pedestrian to be identified, and selecting the security personnel feature with the largest similarity as the security personnel identification result by calculating and comparing the similarity of the final feature of the pedestrian to be identified and the security personnel feature in the database;
in the motion trail matching module, security personnel identification results at different moments are divided into a high-score detection area and a low-score detection area according to a preset similarity threshold, and the current moment positions of security personnel predicted by a security personnel patrol trail set are respectively matched and associated with the high-score detection area and the low-score detection area to obtain security personnel motion trail matching results.
2. The method of claim 1, wherein obtaining pedestrian detection results at different moments in a patrol area according to the pedestrian detection module, and inputting the pedestrian detection results into a pre-constructed feature extraction network for processing, to obtain pedestrian features, comprises:
detecting pedestrians in the patrol area by using a Yolov7 pedestrian detector according to the pedestrian detection module, and obtaining pedestrian detection results at different moments;
inputting the pedestrian detection result into a pre-constructed feature extraction network for processing to obtain pedestrian features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features of the pedestrian detection result
Figure QLYQS_1
The second branch is used for detecting the pedestrians according to the ordinate directionThe result is divided into three equal parts, and the first local feature of the pedestrian detection result of the three equal parts is extracted respectively>
Figure QLYQS_2
Second local feature->
Figure QLYQS_3
And third local feature->
Figure QLYQS_4
。
3. The method of claim 2, wherein in the network training module, clustering the pedestrian features using a density clustering algorithm, and assigning each pedestrian a corresponding pseudo tag according to the clustering result, comprises:
in the network training module, the DbScan density clustering algorithm is adopted for the global features
Figure QLYQS_5
First local feature
Figure QLYQS_6
Second local feature->
Figure QLYQS_7
Third local feature->
Figure QLYQS_8
Clustering, increasing the number of clustering centers by adopting a method of manually checking the clustering result, storing the clustering centers as final clustering results, and distributing a corresponding pseudo tag for each pedestrian according to the final clustering results, wherein the pseudo tag is expressed as
Figure QLYQS_9
；/>
wherein ,
Figure QLYQS_10
respectively represent +.>
Figure QLYQS_11
And the individual person obtains a cluster label according to the global feature, the first local feature, the second local feature and the third local feature.
4. A method according to claim 3, wherein training the feature extraction network according to the optimization function constructed by the pedestrian features and the pseudo tags to obtain a trained feature extraction network comprises:
training the feature extraction network according to the pedestrian features and the optimization function constructed by the pseudo tag by taking the pseudo tag as a data tag, wherein the feature extraction network used in each training is the feature extraction network obtained in the previous training until convergence to obtain a trained feature extraction network; wherein the optimization function is expressed as
Figure QLYQS_12
；
wherein ,
Figure QLYQS_13
representing a ternary loss function, +.>
Figure QLYQS_14
、
Figure QLYQS_15
、
Figure QLYQS_16
and
Figure QLYQS_17
Respectively representing cluster labels obtained according to global features, first local features, second local features and third local features。
5. The method according to claim 1, wherein inputting the pedestrian detection result to be identified into the trained feature extraction network for feature extraction according to the security personnel identification module to obtain the final feature of the pedestrian to be identified, and selecting the security personnel feature with the largest similarity as the security personnel identification result by calculating and comparing the similarity between the final feature of the pedestrian to be identified and the security personnel feature in the database, comprising:
inputting a pedestrian detection result to be identified into a trained feature extraction network according to the security personnel identification module to perform feature extraction to obtain a global feature of the pedestrian to be identified and a local feature to be identified, and splicing the global feature of the pedestrian to be identified and the local feature to be identified to obtain a final feature of the pedestrian to be identified;
selecting nearest security personnel features of the cosine distance as security personnel identification results by calculating and comparing the cosine distance between the final features of the pedestrians to be identified and the security personnel features in the database; the cosine distance nearest represents the maximum similarity between the final characteristics of the pedestrians to be identified and the characteristics of security personnel in the database.
6. The method of claim 1, wherein in the motion trail matching module, security personnel identification results at different moments are divided into a high-score detection area and a low-score detection area according to a preset similarity threshold, and security personnel current moment positions predicted by a security personnel patrol trail set are respectively matched and associated with the high-score detection area and the low-score detection area, so as to obtain security personnel motion trail matching results, wherein the security personnel motion trail matching results are expressed as:
in the motion trail matching module, according to a preset similarity threshold value
Figure QLYQS_18
Dividing security personnel identification results at different moments into high-score detection areas ++>
Figure QLYQS_19
And low-resolution detection region->
Figure QLYQS_20
Wherein, is higher than threshold +.>
Figure QLYQS_21
The security personnel identification result of (2) is classified as +.>
Figure QLYQS_22
Below threshold +.>
Figure QLYQS_23
The identification result of security personnel is classified as +.>
Figure QLYQS_24
；
The security personnel patrol track set T uses the current time position of the security personnel predicted by the Kalman tracker to communicate with the Gao Fenjian area
Figure QLYQS_25
Performing first matching association to obtain a first-stage motion trail matching result; wherein, the high-resolution detection area which is not successfully matched is +.>
Figure QLYQS_26
Stored in the remaining track->
Figure QLYQS_27
In (a) and (b);
the security personnel patrol track set T is respectively connected with the low-resolution detection area by the current time position of the security personnel predicted by the Kalman tracker
Figure QLYQS_28
And remaining track->
Figure QLYQS_29
Performing second matching association to obtain a second-stage motion trail matching result; the method for matching association of the two times is consistent, and security personnel patrol tracks which are not successfully matched in the two times are stored in a mismatch track set +.>
Figure QLYQS_30
In the method, the detection area which is not successfully matched twice is deleted;
and combining the first-stage motion trail matching result and the second-stage motion trail matching result to obtain a security personnel motion trail matching result.
7. The method according to claim 6, wherein the security personnel patrol track set T is used for predicting the current time position of security personnel and the Gao Fenjian area by using a Kalman tracker
Figure QLYQS_31
Performing first matching association to obtain a first-stage motion trail matching result, wherein the first-stage motion trail matching result comprises the following steps of:
calculating the current time position of security personnel predicted by the security personnel patrol track set T through a Kalman tracker and the Gao Fenjian area
Figure QLYQS_32
Setting a cross-over ratio threshold, rejecting matching for the region cross-over ratio smaller than the cross-over ratio threshold, and matching for the region cross-over ratio not smaller than the cross-over ratio threshold to obtain a matching matrix;
and carrying out matching solution on the matching matrix according to the Hungary algorithm to obtain a first-stage motion trail matching result.
8. The method of claim 6, wherein, in performing a matching association, for the set of unmatched tracks
Figure QLYQS_33
Is not matched withMatching tracks, and deleting the unmatched tracks if the existing time exceeds the specified existing time length; otherwise, continuing to store the unmatched track, and when the unmatched track is successfully matched in the subsequent matching association process, collecting the unmatched track from the unmatched track set +.>
Figure QLYQS_34
Transferring to a security personnel patrol track set T; wherein the prescribed length of time of presence is 2 seconds.
9. The utility model provides a security patrol management system based on unsupervised study which characterized in that, security patrol management system based on unsupervised study is through comparing security personnel motion trail matching result and security personnel's regulation patrol orbit the same, judges security personnel's patrol state, realizes security personnel's patrol management, and the system includes:
the pedestrian detection module is used for acquiring pedestrian detection results at different moments in the patrol area, inputting the pedestrian detection results into a pre-constructed feature extraction network for processing, and obtaining pedestrian features; wherein the pedestrian features include global features and local features; the feature extraction network comprises a first branch and a second branch, wherein the first branch is used for extracting global features, and the second branch is used for extracting local features;
the network training module is used for clustering the pedestrian characteristics by adopting a density clustering algorithm, distributing a corresponding pseudo tag for each pedestrian according to a clustering result, and training the characteristic extraction network according to the optimization function constructed by the pedestrian characteristics and the pseudo tags to obtain a trained characteristic extraction network;
the security personnel identification module is used for inputting the pedestrian detection result to be identified into the trained feature extraction network to perform feature extraction to obtain the final feature of the pedestrian to be identified, calculating and comparing the similarity between the final feature of the pedestrian to be identified and the security personnel feature in the database, and selecting the security personnel feature with the maximum similarity as the security personnel identification result;
the motion trail matching module is used for dividing security personnel identification results at different moments into a high-resolution detection area and a low-resolution detection area according to a preset similarity threshold value, and respectively matching and associating the current moment positions of security personnel predicted by the security personnel patrol trail set with the high-resolution detection area and the low-resolution detection area to obtain a security personnel motion trail matching result.
10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 8 when the computer program is executed.
CN202310506915.9A 2023-05-08 2023-05-08 Security patrol management method, system and equipment based on unsupervised learning Pending CN116258751A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310506915.9A CN116258751A (en) 2023-05-08 2023-05-08 Security patrol management method, system and equipment based on unsupervised learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310506915.9A CN116258751A (en) 2023-05-08 2023-05-08 Security patrol management method, system and equipment based on unsupervised learning

Publications (1)

Publication Number Publication Date
CN116258751A true CN116258751A (en) 2023-06-13

Family

ID=86686486

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310506915.9A Pending CN116258751A (en) 2023-05-08 2023-05-08 Security patrol management method, system and equipment based on unsupervised learning

Country Status (1)

Country Link
CN (1) CN116258751A (en)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110942025A (en) * 2019-11-26 2020-03-31 河海大学 Unsupervised cross-domain pedestrian re-identification method based on clustering
CN111967294A (en) * 2020-06-23 2020-11-20 南昌大学 Unsupervised domain self-adaptive pedestrian re-identification method
CN112069940A (en) * 2020-08-24 2020-12-11 武汉大学 Cross-domain pedestrian re-identification method based on staged feature learning
CN112132041A (en) * 2020-09-24 2020-12-25 天津锋物科技有限公司 Community patrol analysis method and system based on computer vision
CN114092964A (en) * 2021-10-19 2022-02-25 杭州电子科技大学 Cross-domain pedestrian re-identification method based on attention guidance and multi-scale label generation
CN114648779A (en) * 2022-03-14 2022-06-21 宁波大学 Unsupervised pedestrian re-identification method based on self-label refined deep learning model
CN114898458A (en) * 2022-04-15 2022-08-12 中国兵器装备集团自动化研究所有限公司 Factory floor number monitoring method, system, terminal and medium based on image processing

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110942025A (en) * 2019-11-26 2020-03-31 河海大学 Unsupervised cross-domain pedestrian re-identification method based on clustering
CN111967294A (en) * 2020-06-23 2020-11-20 南昌大学 Unsupervised domain self-adaptive pedestrian re-identification method
CN112069940A (en) * 2020-08-24 2020-12-11 武汉大学 Cross-domain pedestrian re-identification method based on staged feature learning
CN112132041A (en) * 2020-09-24 2020-12-25 天津锋物科技有限公司 Community patrol analysis method and system based on computer vision
CN114092964A (en) * 2021-10-19 2022-02-25 杭州电子科技大学 Cross-domain pedestrian re-identification method based on attention guidance and multi-scale label generation
CN114648779A (en) * 2022-03-14 2022-06-21 宁波大学 Unsupervised pedestrian re-identification method based on self-label refined deep learning model
CN114898458A (en) * 2022-04-15 2022-08-12 中国兵器装备集团自动化研究所有限公司 Factory floor number monitoring method, system, terminal and medium based on image processing

Similar Documents

Publication Publication Date Title
CN112200111B (en) Global and local feature fused occlusion robust pedestrian re-identification method
Luo et al. Towards efficient and objective work sampling: Recognizing workers' activities in site surveillance videos with two-stream convolutional networks
CN110569701B (en) Computer-implemented vehicle damage assessment method and device
CN108509859B (en) Non-overlapping area pedestrian tracking method based on deep neural network
CN112380870A (en) User intention analysis method and device, electronic equipment and computer storage medium
KR101917354B1 (en) System and Method for Multi Object Tracking based on Reliability Assessment of Learning in Mobile Environment
CN111275060A (en) Recognition model updating processing method and device, electronic equipment and storage medium
CN111428619B (en) Three-dimensional point cloud head attitude estimation system and method based on ordered regression and soft labels
CN111476315A (en) Image multi-label identification method based on statistical correlation and graph convolution technology
CN110991513A (en) Image target recognition system and method with human-like continuous learning capability
CN110781818B (en) Video classification method, model training method, device and equipment
US20220207741A1 (en) Methods, devices, apparatuses and storage media of detecting correlated objects invovled in images
CN111738074B (en) Pedestrian attribute identification method, system and device based on weak supervision learning
CN112749653A (en) Pedestrian detection method, device, electronic equipment and storage medium
Wang et al. Neural koopman pooling: Control-inspired temporal dynamics encoding for skeleton-based action recognition
CN113780145A (en) Sperm morphology detection method, sperm morphology detection device, computer equipment and storage medium
CN114255377A (en) Differential commodity detection and classification method for intelligent container
CN112528903B (en) Face image acquisition method and device, electronic equipment and medium
US20220300774A1 (en) Methods, apparatuses, devices and storage media for detecting correlated objects involved in image
Nath et al. Deep learning models for content-based retrieval of construction visual data
Duong et al. Helmet rule violation detection for motorcyclists using a custom tracking framework and advanced object detection techniques
CN111241961B (en) Face detection method and device and electronic equipment
CN113269038A (en) Multi-scale-based pedestrian detection method
CN115049836B (en) Image segmentation method, device, equipment and storage medium
CN116453226A (en) Human body posture recognition method and device based on artificial intelligence and related equipment

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