CN111339879B - Weapon room single person room entering monitoring method and device - Google Patents

Weapon room single person room entering monitoring method and device Download PDF

Info

Publication number
CN111339879B
CN111339879B CN202010102238.0A CN202010102238A CN111339879B CN 111339879 B CN111339879 B CN 111339879B CN 202010102238 A CN202010102238 A CN 202010102238A CN 111339879 B CN111339879 B CN 111339879B
Authority
CN
China
Prior art keywords
room
model
weapon
representing
camera
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.)
Active
Application number
CN202010102238.0A
Other languages
Chinese (zh)
Other versions
CN111339879A (en
Inventor
徐国
江瀚澄
熊忠元
苏丹
张新选
徐斌
曹振武
徐孝文
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.)
Anhui Lingyun Iot Technology Co ltd
Original Assignee
Anhui Lingyun Iot 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 Anhui Lingyun Iot Technology Co ltd filed Critical Anhui Lingyun Iot Technology Co ltd
Priority to CN202010102238.0A priority Critical patent/CN111339879B/en
Publication of CN111339879A publication Critical patent/CN111339879A/en
Application granted granted Critical
Publication of CN111339879B publication Critical patent/CN111339879B/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
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • 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
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • 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)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a weapon room single person room entering monitoring method and device, wherein the method comprises the following steps: the camera collects the video stream of weapon room of a certain armed police team; converting the video stream acquired by the camera into a frame image, screening out partial images and calibrating the images; inputting the calibrated image into a YOLO v3 model, and training the YOLO v3 model to obtain a personnel identification model; the personnel identification model is utilized to carry out weapon room single person room entering monitoring, and the detection result of the model is input into the linkage model; the linkage detection model calculates the maximum value of uploading data of each camera every second, and if the maximum value calculated by any camera is 1, the linkage detection model judges that the weapon room has single-person room entering behavior at the moment, and if the linkage detection model judges that the weapon room has single-person room entering behavior in m seconds, an alarm is sent to the weapon room monitoring system; the invention has the advantages that: the detection accuracy is high, and the method is suitable for a weapon room single person marketing alarm scene.

Description

Weapon room single person room entering monitoring method and device
Technical Field
The invention relates to the field of weapon room personnel monitoring, in particular to a weapon room single person room entering monitoring method and device.
Background
The army police 'army weapon room management regulation' clearly indicates that the weapon room should have a main department director, and the army personnel is particularly responsible for implementing a double lock and double warehouse entry and exit system. The conventional practice is as follows: the gun taker starts a gate and then enters a room to take a gun through the authentication of a double face or fingerprint of a weapon room entrance guard under the accompany of a ordnance or a main department; and installing a high-definition camera at the secret-related place of the army, and remotely monitoring the dynamic state of the place in real time by an operator on duty. The safety hidden danger exists in the warehouse-in and warehouse-out and manual monitoring modes, and the report missing is easy to occur, for example: after the army personnel and the duty trunk part open the weapon room gate, the army personnel are not accompanied to enter the room for picking up the gun; the ordnance enters with the gun taker, but the two are distributed in different rooms. By applying the intelligent video analysis technology, the real-time detection and automatic reporting of the weapon room single person warehouse entry can be realized, and the detection effect of abnormal behaviors is effectively improved.
Chinese patent publication No. CN105719368A discloses a personnel detecting system and a personnel detecting method, the personnel detecting system includes a thermal sensor and an indoor device. The thermal sensor extracts a thermal image of the indoor space. The personnel detection system switches among a personnel entering detection mode, a first personnel leaving detection mode corresponding to the closing of the indoor equipment and a second personnel leaving detection mode corresponding to the opening of the indoor equipment, wherein the personnel detection system executes a personnel entering detection program according to the thermal image in the personnel entering detection mode; in the first person leaving detection mode and the second person leaving detection mode, a person leaving detection program is executed based on the thermal image. According to the invention, the thermal image is used for detecting the personnel, and the detection mode is switched according to the state of the indoor equipment, so that the erroneous judgment of the entering/leaving state of the personnel can be effectively avoided.
Chinese patent publication No. CN109740522A discloses a personnel detection method, device, equipment and medium. The method comprises the following steps: performing convolutional neural network training on a radio frequency signal thermal pattern book containing human body posture information to generate a radio frequency signal identification model; the radiofrequency signal heat map sample is generated by transmitting radiofrequency signals to a human body in advance and drawing the radiofrequency signals according to reflection signals of the human body on the radiofrequency signals; the method comprises the steps of obtaining a target radio frequency signal heat map under a target scene by transmitting radio frequency signals, inputting the target radio frequency signal heat map into a radio frequency signal identification model to detect whether personnel exist in the target scene or not, and obtaining a detection result. The method can relatively improve the accuracy of personnel detection in the space and relatively ensure the overall reliability of personnel detection. In addition, the invention also provides a personnel detection device, equipment and medium, and the beneficial effects are the same as those described above.
The thermal image or the radio frequency signal heat map is used for realizing personnel detection, and the defects are that the thermal image and the radio frequency signal heat map lose too much personnel details, and only the personnel detection is realized, that is, whether people exist or not can be identified, the number of people cannot be identified, and then the single person entering room alarm cannot be realized in the prior art. Because the technology loses too many figure details, the identification accuracy is poor, and if the technology is directly used for a weapon room single person to enter a room monitoring system, the phenomenon of missing report or false report can often occur.
Disclosure of Invention
The technical problem to be solved by the invention is that the personnel detection method in the prior art is not suitable for the weapon room single person marketing alarm scene.
The invention solves the technical problems by the following technical means: a weapon room single person-in-room monitoring method, the method comprising:
step one: the camera collects the video stream of weapon room of a certain armed police team;
step two: converting the video stream acquired by the camera into a frame image, screening out partial images and calibrating the images;
step three: inputting the calibrated image into a YOLO v3 model, and training the YOLO v3 model to obtain a personnel identification model;
step four: the personnel identification model is utilized to carry out weapon room single person room entering monitoring, and the detection result of the model is input into the linkage model;
step five: the linkage detection model receives and stores the number of people data uploaded by the personnel identification model in real time, calculates the maximum value of the data uploaded by each camera every second, and only if the maximum value calculated by any camera is 1, the linkage detection model judges that the weapon room has single room entering behavior at the moment, and if the linkage detection model judges that the weapon room has single room entering behavior within m seconds continuously, an alarm is sent to the weapon room monitoring system.
The video stream of the weapon room is collected through the camera, the calibrated image is input into the YOLO v3 model, the robustness and the identification accuracy of the YOLO v3 model are high, the alarm is realized through the cooperation of the personnel identification model and the linkage detection model, the problem of missing report and hidden report caused by the objective condition and subjective thinking of a person is solved, meanwhile, the maximum value calculated by any camera is 1, the linkage detection model judges that the weapon room has single entering behavior at the moment, and if the linkage detection model judges that the weapon room has single entering behavior within m seconds, the alarm is sent to the weapon room monitoring system, so that the weapon room monitoring system is suitable for the scene of single entering monitoring of the weapon room.
Preferably, the first step includes: each room of weapon room all installs a camera, and the subaerial mark that corresponds at the camera shooting dead angle is forbidden the district, and the armed police fighter of different statures gets into the weapon room of waiting to monitor and walk at will with different uniform, and video acquisition video stream is recorded to many cameras.
Preferably, in the second step, a frame image of the video stream is acquired by using an OPEN CV.
Preferably, in the second step, the yolo_mark tool is used to calibrate the image.
Preferably, in the third step, the YOLO v3 model is trained using a yolov3.weights pre-training model.
Preferably, the loss function of the YOLO v3 model is
Figure BDA0002387245480000041
wherein ,
Figure BDA0002387245480000042
Figure BDA0002387245480000043
λ coord representing the scaling factor of the coordinate loss function; s is S 2 Representing the total grid number of the training single picture; i represents the ith grid; j represents the j-th anchor block; b represents the number of anchor blocks per grid; obj represents the target object for which the anchor block is responsible; smoothL 1 () Represent smoothL 1 Loss function, diff represents coordinate loss function, L 1 (diff) smoothL representing diff 1 A loss value; c (C) i Representing the output confidence;
Figure BDA0002387245480000044
representing the true confidence; lambda (lambda) noobj Representing a scaling factor of an anchor not responsible for predicting coordinates; class represents all categories; c represents each object; p is p i (c) Representing the output probability of each object; />
Figure BDA0002387245480000045
Representing the real probability of each object; x is x i Representing the abscissa of the central point of the predicted frame; />
Figure BDA0002387245480000046
Representing the abscissa of the center point of the actual frame; y is i Representing the ordinate of the central point of the predicted frame; />
Figure BDA0002387245480000047
Representing the ordinate of the center point of the actual frame; omega i Representing a predicted border width; />
Figure BDA0002387245480000048
Representing the actual frame width; h is a i Representing a predicted bezel height; />
Figure BDA0002387245480000049
Representing the actual bezel height.
Preferably, m has a value in the range of 5< m <7.
The invention also provides a weapon room single person room entering monitoring device, which comprises:
the acquisition module is used for acquiring the weapon room video stream of a certain armed police;
the calibration module is used for converting the video stream acquired by the camera into a frame image, screening out partial images and calibrating the images;
the personnel identification model acquisition module is used for inputting the calibrated image into the YOLO v3 model, and training the YOLO v3 model to obtain a personnel identification model;
the detection module is used for carrying out weapon room single person room entering monitoring by using the personnel identification model, and inputting the detection result of the model into the linkage model;
the alarm module is used for receiving and storing the number of people data uploaded by the personnel identification model in real time by the linkage detection model, calculating the maximum value of the data uploaded by each camera of the previous second every second, judging that the weapon room has single room entering behavior at the moment by the linkage detection model as long as the maximum value calculated by any camera is 1, and sending an alarm to the weapon room supervision system if the single room entering behavior of the weapon room is judged by the linkage detection model within m seconds.
Preferably, the acquisition module is further configured to: each room of weapon room all installs a camera, and the subaerial mark that corresponds at the camera shooting dead angle is forbidden the district, and the armed police fighter of different statures gets into the weapon room of waiting to monitor and walk at will with different uniform, and video acquisition video stream is recorded to many cameras.
Preferably, in the calibration module, an OPEN CV is used to acquire a frame image of the video stream.
Preferably, in the calibration module, a yolo_mark tool is used for calibrating the image.
Preferably, in the person recognition model acquisition module, a YOLO v3 model is trained using a yolov3.weights pre-training model.
Preferably, the loss function of the YOLO v3 model is
Figure BDA0002387245480000061
/>
wherein ,
Figure BDA0002387245480000062
Figure BDA0002387245480000063
λ coord representing the scaling factor of the coordinate loss function; s is S 2 Representing the total grid number of the training single picture; i represents the ith grid; j represents the j-th anchor block; b represents the number of anchor blocks per grid; obj represents the target object for which the anchor block is responsible; smoothL 1 () Represent smoothL 1 Loss function, diff represents coordinate loss function, L 1 (diff) smoothL representing diff 1 A loss value; c (C) i Representing the output confidence;
Figure BDA0002387245480000064
representing the true confidence; lambda (lambda) noobj Representing a scaling factor of an anchor not responsible for predicting coordinates; class represents all categories; c represents each object; p is p i (c) Representing the output probability of each object; />
Figure BDA0002387245480000065
Representing the real probability of each object; x is x i Representing the abscissa of the central point of the predicted frame; />
Figure BDA0002387245480000066
Representing the abscissa of the center point of the actual frame; y is i Representing the ordinate of the central point of the predicted frame; />
Figure BDA0002387245480000067
Representing the ordinate of the center point of the actual frame; omega i Representing a predicted border width; />
Figure BDA0002387245480000068
Representing the actual frame width; h is a i Representing a predicted bezel height; />
Figure BDA0002387245480000069
Representing the actual bezel height.
Preferably, m has a value in the range of 5< m <7.
The invention has the advantages that:
(1) The video stream of the weapon room is collected through the camera, the calibrated image is input into the YOLO v3 model, the robustness and the identification accuracy of the YOLO v3 model are high, the alarm is realized through the cooperation of the personnel identification model and the linkage detection model, the problem of missing report and hidden report caused by the objective condition and subjective thinking of a person is solved, meanwhile, the maximum value calculated by any camera is 1, the linkage detection model judges that the weapon room has single entering behavior at the moment, and if the linkage detection model judges that the weapon room has single entering behavior within m seconds, the alarm is sent to the weapon room monitoring system, so that the weapon room monitoring system is suitable for the scene of single entering monitoring of the weapon room.
(2) The invention can be widely applied to monitoring security rooms, archives, combat command rooms and the like, checking whether personnel are in place or not and whether abnormal behaviors exist or not, and has wide application range.
(3) The invention improves the loss function of the YOLO v3 model, and is composed of smoothL 1 (x) The function derives x, which is the reciprocal value 1 when |x| is equal to or greater than 1, and the derivative x when |x| is less than 1. At the early stage of training, smoothL thereof 1 (x) The value of the derivative x is 1, so that the initial stability of training is ensured; later in training, the prediction frame differs less from the real object frame (x value is less than 1) smoothL 1 (x) The diff advantage is inherited, so that the model is easy to converge to higher precision, and the robustness and the recognition accuracy of the YOLOv3 model are enhanced.
Drawings
FIG. 1 is a flow chart of a method for monitoring a weapon room by a single person;
FIG. 2 is a system architecture diagram of a weapon room single person-in-room monitoring method according to an embodiment of the present invention;
fig. 3 is a flow chart of a linkage detection model of a weapon room single person room entering monitoring method according to an embodiment of the invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Example 1
As shown in fig. 1 and 2, a weapon room single person room entering monitoring method includes:
step S1: the camera collects the video stream of weapon room of a certain armed police team; the method comprises the following specific steps: the weapon room typically has a plurality of rooms, each of which houses a camera. Because there is a dead angle of a small circle under the camera, should install the camera in getting the corner that firearms can not pass, just so avoided because the condition that single entering room leaks the report that the dead angle leads to of shooing to mark an obvious forbidden area on the subaerial that the camera shooted the dead angle to correspond, be equivalent to marking out the dead angle, set the dead angle as forbidden area, make personnel can not appear at the dead angle, prevent the emergence of false alarm phenomenon. The camera has two functions, namely, the camera acquires pictures for secondary training of the personnel identification model, and the camera acquires real-time monitoring videos of a plurality of rooms of the weapon room when the system operates. The video stream is mainly obtained by the steps that armed police fighters of different sizes enter a weapon room to be monitored to walk at will with different uniforms, and the video stream is obtained by recording multiple cameras.
Step S2: converting the video stream acquired by the camera into a frame image, screening out partial images and calibrating the images; wherein the frame images of the video stream are acquired using OPEN CV. The image was calibrated using the yolo_mark tool.
Step S3: inputting the calibrated image into a YOLO v3 model, and training the YOLO v3 model to obtain a personnel identification model; the models related to the embodiment of the invention are all prior art models, and only the loss function of the YOLO v3 model is adjusted, so that detailed processes of building each model and training the model are not repeated. Wherein the YOLO v3 model is trained using the yolov3.Weights pre-training model. The personnel identification model detects personnel on the frame image of the real-time monitoring video stream, the model outputs coordinates of the target frame, the number of the output coordinates is calculated, and the number of the output coordinates is used as a detection result to be sent to the linkage detection model. The YOLO v3 model uses a dark net-53 convolution model as a network structure for basic feature extraction, the dark net-53 network has 74 layers, wherein 53 convolution layers (the convolution layers are mainly 1x1 and 3x3 in size, each convolution layer is followed by one BN layer and one LeakyReLU layer), the rest is a residual layer, and residual modules are added into the network, so that the gradient problem of the deep network is solved, and each residual module is formed by two convolution layers and one shortcut connection (shortcut connections).
Since the YOLO algorithm of the original edition detects multiple targets simultaneously, and only 'people' need to be detected in the application scene of the invention, the model number of the model number in the configuration file yolo.cfg is improved according to the characteristic that the aspect ratio of the human is 0.4, then pre-training super-parameters are imported, and the data set in the step S1 is used for training the YOLO v3 model to obtain a final personnel identification model. Wherein yolo.cfg modification parameters are as follows:
anchors＝16,16,16,32,32,32,64,32,32,64,64,64,128,64,64,128,128,128,256,128,128,256,256,256。
the invention also improves the loss function, and the loss function of the YOLO v3 model is as follows:
Figure BDA0002387245480000091
/>
wherein ,
Figure BDA0002387245480000092
Figure BDA0002387245480000093
λ coord representing the scaling factor of the coordinate loss function; s is S 2 Representing the total grid number of the training single picture; i represents the ith grid; j represents the j-th anchor block; b represents the number of anchor blocks per grid; obj represents the target object for which the anchor block is responsible; smoothL 1 () Represent smoothL 1 Loss function, diff represents coordinate loss function, L 1 (diff) smoothL representing diff 1 A loss value; c (C) i Representing the output confidence;
Figure BDA0002387245480000102
representing the true confidence; lambda (lambda) noobj Representing a scaling factor of an anchor not responsible for predicting coordinates; class represents all categories; c represents each object; p is p i (c) Representing the output probability of each object; />
Figure BDA0002387245480000103
Representing the real probability of each object; x is x i Representing the abscissa of the central point of the predicted frame; />
Figure BDA0002387245480000104
Representing the abscissa of the center point of the actual frame; y is i Representing the ordinate of the central point of the predicted frame; />
Figure BDA0002387245480000105
Representing the ordinate of the center point of the actual frame; omega i Representing a predicted border width; />
Figure BDA0002387245480000106
Representing the actual frame width; h is a i Representing a predicted bezel height; />
Figure BDA0002387245480000107
Representing the actual bezel height.
The derivative of the original frame loss function diff is 2x, and in the initial stage of training, the difference between a predicted frame and a real object frame is too large (the value of x is far greater than 1), and the derivative is also large. And smoothL 1 (x) And (3) deriving:
Figure BDA0002387245480000101
from smoothL 1 (x) The function derives x, which is the reciprocal value 1 when |x| is equal to or greater than 1, and the derivative x when |x| is less than 1. At the early stage of training, smoothL thereof 1 (x) The value of the derivative x is 1, so that the initial stability of training is ensured; later in training, the prediction frame differs less from the real object frame (x value is less than 1) smoothL 1 (x) Inherit diff advantages, and enable the model to be easy to converge to higher precision.
The method improves the loss function of the YOLO v3 model, and enhances the robustness and the identification accuracy of the YOLO v3 model. In actual tests, the recall rate identified by the method is close to 98.6%, the accuracy rate is up to more than 97.3%, and the method is obviously improved compared with the original yolov3 algorithm model in the prior art.
Step S4: the personnel identification model is utilized to carry out weapon room single person room entering monitoring, and the detection result of the model is input into the linkage model;
step S5: the linkage detection model receives and stores the number of people data uploaded by the personnel identification model in real time, calculates the maximum value of the data uploaded by each camera every second, and only needs that the maximum value calculated by any camera is 1, the linkage detection model judges that the weapon room has single room entering behavior at the moment, and the probability of false alarm occurrence of each frame of picture is 0.1 after the strategy is used, assuming that each camera acquires 20 frames of images every second, and the probability of false alarm occurrence of the system is 0.1:
p (false alarm) =0.1 20
If the linkage detection model judges that the weapon room has single person room entering behavior within m seconds continuously, an alarm is sent to the weapon room supervision system, and the value range of m is 5< m <7, and m is 6 in the embodiment of the invention. Assuming that the probability of a model false alarm occurring every second is 0.05, the probability of a system false alarm occurring after using this strategy:
p (false alarm) =0.05 6
The two strategies are combined to greatly reduce the probability of false alarm and frequent alarm problems.
Assuming that the system has two monitoring cameras, initializing an abnormal tag E to be 0, receiving the number of people data D1 and D2 of cameras C1 and C2 detected by a YOLO v3 detection model by a linkage monitoring model S, storing the D1 and D2 into memories P1 and P2, calculating max (P1) and max (P2) every second, then clearing the memory data, performing OR operation on max (P1) and max (P2), if the result is 1, E=E+1, otherwise resetting E to be 0, and warning the weapon room supervision system by the model when E=6. The linkage detection model flow chart is shown in fig. 3.
According to the technical scheme, the camera is used for collecting video streams of the weapon room, calibrated images are input into the YOLO v3 model, the robustness and the identification accuracy of the YOLO v3 model are high, the personnel identification model and the linkage detection model are matched together to realize alarming, the problem of missing report and hidden report caused by objective conditions and subjective thinking of a person is solved, meanwhile, the maximum value calculated by any camera is 1, the linkage detection model judges that single person entering behavior of the weapon room occurs at the moment, and if the linkage detection model judges that single person entering behavior of the weapon room occurs within m seconds, the linkage detection model sends alarming to the weapon room monitoring system, so that the method is suitable for scenes of single person entering monitoring of the weapon room.
Example 2
Corresponding to embodiment 1 of the present invention, embodiment 2 of the present invention further provides a weapon room single person entering room monitoring device, the device comprising:
the acquisition module is used for acquiring the weapon room video stream of a certain armed police;
the calibration module is used for converting the video stream acquired by the camera into a frame image, screening out partial images and calibrating the images;
the personnel identification model acquisition module is used for inputting the calibrated image into the YOLO v3 model, and training the YOLO v3 model to obtain a personnel identification model;
the detection module is used for carrying out weapon room single person room entering monitoring by using the personnel identification model, and inputting the detection result of the model into the linkage model;
the alarm module is used for receiving and storing the number of people data uploaded by the personnel identification model in real time by the linkage detection model, calculating the maximum value of the data uploaded by each camera of the previous second every second, judging that the weapon room has single room entering behavior at the moment by the linkage detection model as long as the maximum value calculated by any camera is 1, and sending an alarm to the weapon room supervision system if the single room entering behavior of the weapon room is judged by the linkage detection model within m seconds.
Specifically, the acquisition module is further configured to: each room of weapon room all installs a camera, and the subaerial mark that corresponds at the camera shooting dead angle is forbidden the district, and the armed police fighter of different statures gets into the weapon room of waiting to monitor and walk at will with different uniform, and video acquisition video stream is recorded to many cameras.
Specifically, in the calibration module, an OPEN CV is used to obtain a frame image of the video stream.
Specifically, in the calibration module, a yolo_mark tool is used for calibrating the image.
Specifically, in the personnel identification model acquisition module, a YOLO v3 model is trained by using a yolov3.weights pre-training model.
Preferably, the loss function of the YOLO v3 model is
Figure BDA0002387245480000131
/>
wherein ,
Figure BDA0002387245480000132
Figure BDA0002387245480000133
λ coord representing the scaling factor of the coordinate loss function; s is S 2 Representing the total grid number of the training single picture; i represents the ith grid; j represents the j-th anchor block; b represents the number of anchor blocks per grid; obj represents the target object for which the anchor block is responsible; smoothL 1 () Represent smoothL 1 Loss function, diff represents coordinate loss function, L 1 (diff) smoothL representing diff 1 A loss value; c (C) i Representing the output confidence;
Figure BDA0002387245480000134
representing the true confidence; lambda (lambda) noobj Representing a scaling factor of an anchor not responsible for predicting coordinates; class represents all categories; c represents each object; p is p i (c) Representing the output probability of each object; />
Figure BDA0002387245480000135
Representing the real probability of each object; x is x i Representing the abscissa of the central point of the predicted frame; />
Figure BDA0002387245480000136
Representing the abscissa of the center point of the actual frame; y is i Representing the ordinate of the central point of the predicted frame; />
Figure BDA0002387245480000139
Representing the ordinate of the center point of the actual frame; omega i Representing a predicted border width; />
Figure BDA0002387245480000137
Representing the actual frame width; h is a i Representing a predicted bezel height; />
Figure BDA0002387245480000138
Representing the actual bezel height.
Specifically, the value range of m is 5< m <7.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A weapon room single person-in-room monitoring method, the method comprising:
step one: the camera collects the video stream of weapon room of a certain armed police team;
step two: converting the video stream acquired by the camera into a frame image, screening out partial images and calibrating the images;
step three: inputting the calibrated image into a YOLO v3 model, and training the YOLO v3 model to obtain a personnel identification model;
step four: the personnel identification model is utilized to carry out weapon room single person room entering monitoring, and the detection result of the model is input into the linkage model;
step five: the linkage detection model receives and stores the number of people data uploaded by the personnel identification model in real time, calculates the maximum value of the data uploaded by each camera every second, and only if the maximum value calculated by any camera is 1, the linkage detection model judges that the weapon room has single room entering behavior at the moment, and if the linkage detection model judges that the weapon room has single room entering behavior within m seconds continuously, an alarm is sent to the weapon room monitoring system.
2. The method of claim 1, wherein the first step comprises: each room of weapon room all installs a camera, and the subaerial mark that corresponds at the camera shooting dead angle is forbidden the district, and the armed police fighter of different statures gets into the weapon room of waiting to monitor and walk at will with different uniform, and video acquisition video stream is recorded to many cameras.
3. The method for monitoring the entrance of a weapon room by a single person according to claim 1, wherein in the second step, the OPEN CV is used to acquire the frame image of the video stream.
4. The method for monitoring the entrance of a weapon room by a single person according to claim 1, wherein in the second step, the image is calibrated by using a yolo_mark tool.
5. The method of claim 1, wherein in the third step, YOLO v3 model is trained using yolov3.Weights pre-training model.
6. The weapon room single person room monitoring method as claimed in claim 1, wherein the loss function of the YOLO v3 model is
Figure FDA0002387245470000021
wherein ,
Figure FDA0002387245470000022
Figure FDA0002387245470000023
λ coord representing the scaling factor of the coordinate loss function; s is S 2 Representing the total grid number of the training single picture; i represents the ith grid; j represents the j-th anchor block; b represents the number of anchor blocks per grid; obj represents the target object for which the anchor block is responsible; smoothL 1 () Represent smoothL 1 Loss function, diff represents coordinate loss function, L 1 (diff) smoothL representing diff 1 A loss value; c (C) i Representing the output confidence;
Figure FDA0002387245470000024
representing the true confidence; lambda (lambda) noobj Representing a scaling factor of an anchor not responsible for predicting coordinates; class represents all categories; c represents each object; p is p i (c) Representing the output probability of each object; />
Figure FDA0002387245470000025
Representing the real probability of each object; x is x i Representing the abscissa of the central point of the predicted frame; />
Figure FDA0002387245470000026
Representing the abscissa of the center point of the actual frame; y is i Representing the ordinate of the central point of the predicted frame; />
Figure FDA0002387245470000027
Representing the ordinate of the center point of the actual frame; omega i Representing a predicted border width; />
Figure FDA0002387245470000028
Representing the actual frame width; h is a i Representing a predicted bezel height; />
Figure FDA0002387245470000029
Representing the actual bezel height.
7. The weapon room single person-in-room monitoring method according to claim 1, wherein the value range of m is 5< m <7.
8. A weapon room single person room entering monitoring device, characterized in that the device comprises:
the acquisition module is used for acquiring the weapon room video stream of a certain armed police;
the calibration module is used for converting the video stream acquired by the camera into a frame image, screening out partial images and calibrating the images;
the personnel identification model acquisition module is used for inputting the calibrated image into the YOLO v3 model, and training the YOLO v3 model to obtain a personnel identification model;
the detection module is used for carrying out weapon room single person room entering monitoring by using the personnel identification model, and inputting the detection result of the model into the linkage model;
the alarm module is used for receiving and storing the number of people data uploaded by the personnel identification model in real time by the linkage detection model, calculating the maximum value of the data uploaded by each camera of the previous second every second, judging that the weapon room has single room entering behavior at the moment by the linkage detection model as long as the maximum value calculated by any camera is 1, and sending an alarm to the weapon room supervision system if the single room entering behavior of the weapon room is judged by the linkage detection model within m seconds.
9. The weapon room single entry monitoring device of claim 8, wherein the acquisition module is further configured to: each room of weapon room all installs a camera, and the subaerial mark that corresponds at the camera shooting dead angle is forbidden the district, and the armed police fighter of different statures gets into the weapon room of waiting to monitor and walk at will with different uniform, and video acquisition video stream is recorded to many cameras.
10. The weapon room single person-in-room monitoring device according to claim 8, wherein the calibration module acquires frame images of the video stream using OPEN CV.
CN202010102238.0A 2020-02-19 2020-02-19 Weapon room single person room entering monitoring method and device Active CN111339879B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010102238.0A CN111339879B (en) 2020-02-19 2020-02-19 Weapon room single person room entering monitoring method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010102238.0A CN111339879B (en) 2020-02-19 2020-02-19 Weapon room single person room entering monitoring method and device

Publications (2)

Publication Number Publication Date
CN111339879A CN111339879A (en) 2020-06-26
CN111339879B true CN111339879B (en) 2023-06-02

Family

ID=71181654

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010102238.0A Active CN111339879B (en) 2020-02-19 2020-02-19 Weapon room single person room entering monitoring method and device

Country Status (1)

Country Link
CN (1) CN111339879B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114025197A (en) * 2021-11-01 2022-02-08 浙江力石科技股份有限公司 Anti-counterfeiting encryption method based on video stream

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017177902A1 (en) * 2016-04-14 2017-10-19 平安科技（深圳）有限公司 Video recording method, server, system, and storage medium
CN109711320A (en) * 2018-12-24 2019-05-03 兴唐通信科技有限公司 A kind of operator on duty's unlawful practice detection method and system
CN110321853A (en) * 2019-07-05 2019-10-11 杭州巨骐信息科技股份有限公司 Distribution cable external force damage prevention system based on video intelligent detection

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017177902A1 (en) * 2016-04-14 2017-10-19 平安科技（深圳）有限公司 Video recording method, server, system, and storage medium
CN109711320A (en) * 2018-12-24 2019-05-03 兴唐通信科技有限公司 A kind of operator on duty's unlawful practice detection method and system
CN110321853A (en) * 2019-07-05 2019-10-11 杭州巨骐信息科技股份有限公司 Distribution cable external force damage prevention system based on video intelligent detection

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
刘君 ; 谢颖华 ; .智能视频监控系统中改进YOLO算法的实现.信息技术与网络安全.2019,(04),全文. *

Also Published As

Publication number Publication date
CN111339879A (en) 2020-06-26

Similar Documents

Publication Publication Date Title
CN109461168B (en) Target object identification method and device, storage medium and electronic device
US20240062518A1 (en) Method, apparatus, computer device and storage medium for detecting objects thrown from height
CN101123722B (en) Panorama video intelligent monitoring method and system
US9520040B2 (en) System and method for real-time 3-D object tracking and alerting via networked sensors
KR102195706B1 (en) Method and Apparatus for Detecting Intruder
CN101119482B (en) Overall view monitoring method and apparatus
US7751647B2 (en) System and method for detecting an invalid camera in video surveillance
JP5459674B2 (en) Moving object tracking system and moving object tracking method
Bhadwal et al. Smart border surveillance system using wireless sensor network and computer vision
KR102058452B1 (en) IoT Convergence Intelligent Video Analysis Platform System
CN109326073A (en) A kind of wireless sensor network safety defense monitoring system based on multipoint positioning
CN103929592A (en) All-dimensional intelligent monitoring equipment and method
CN113963301A (en) Space-time feature fused video fire and smoke detection method and system
CN114724330A (en) Implementation method of self-adaptive mode switching multi-channel video fire real-time alarm system
CN111882810A (en) Fire identification and early warning method and system
CN111339879B (en) Weapon room single person room entering monitoring method and device
CN109887211A (en) Human body method for detecting
CN114120171A (en) Fire smoke detection method, device and equipment based on video frame and storage medium
CN209433517U (en) It is a kind of based on more flame images and the fire identification warning device for combining criterion
KR20240028290A (en) Device and method for detecting wildfire
Munawar et al. Fire detection through Image Processing; A brief overview
CN113225525B (en) Indoor monitoring method and system
CN115841730A (en) Video monitoring system and abnormal event detection method
KR102500853B1 (en) Fight Situation Monitering Method Based on Lighted Deep Learning and System thereof
KR20210086536A (en) Method for monitoring fire and providing fire alarm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant