CN117576719A - Method for detecting illegal behaviors in operation scene - Google Patents

Method for detecting illegal behaviors in operation scene Download PDF

Info

Publication number
CN117576719A
CN117576719A CN202210938993.1A CN202210938993A CN117576719A CN 117576719 A CN117576719 A CN 117576719A CN 202210938993 A CN202210938993 A CN 202210938993A CN 117576719 A CN117576719 A CN 117576719A
Authority
CN
China
Prior art keywords
picture
loss
target
detecting
behavior
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
CN202210938993.1A
Other languages
Chinese (zh)
Inventor
汪依帆
龙飞
王海洋
郭海伟
陈凤仙
马红卫
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.)
Information Management Center Of Zhongyuan Oilfield Branch Of Sinopec
China Petroleum and Chemical Corp
Original Assignee
Information Management Center Of Zhongyuan Oilfield Branch Of Sinopec
China Petroleum and Chemical Corp
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 Information Management Center Of Zhongyuan Oilfield Branch Of Sinopec, China Petroleum and Chemical Corp filed Critical Information Management Center Of Zhongyuan Oilfield Branch Of Sinopec
Priority to CN202210938993.1A priority Critical patent/CN117576719A/en
Publication of CN117576719A publication Critical patent/CN117576719A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • 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
    • 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
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • 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
    • 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
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/44—Event detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • 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)
  • Multimedia (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Image Analysis (AREA)

Abstract

The invention belongs to the technical field of target detection, and particularly relates to a method for detecting illegal behaviors in a working scene. The invention improves the loss function of the YOLOv3 algorithm model and adopts L GIoU The loss function replaces the IOU loss function of the traditional YOLOv3 algorithm, andthe improved YOLOv3 algorithm model is applied to on-site offence behavior detection. The invention introduces the minimum external frame on the basis of the IOU characteristic, so as to solve the problem that the loss is equal to 0 when the detection frame and the real frame are not overlapped, thereby improving the accuracy of target detection, being beneficial to identifying smaller targets, especially facing the problems of interference, illumination change and background interference caused by object shielding, effectively improving the detection effect and being beneficial to guaranteeing the safety of drilling operation.

Description

Method for detecting illegal behaviors in operation scene
Technical Field
The invention belongs to the technical field of target detection, and particularly relates to a method for detecting illegal behaviors in a working scene.
Background
The industrial site has the characteristics of high safety risk, numerous contained personnel and wide operation range, new conditions and new risks are continuously emerging, the difficulty of controlling the new conditions and the new risks is high, and accidents occur. In the face of such situation, simply rely on the manpower to carry out safety supervision and on-the-spot management and control, have the problem that supervision inefficiency, human cost are high, supervision precision is low, and this just puts forward new requirement to safety management work.
In recent years, target detection technology based on computer vision has been gradually applied to the field of petroleum drilling engineering. Object Detection (Object Detection) is one of the most hot research directions in computer vision, which is to determine the position of an Object in an input image and identify the category to which the Object in the input image belongs. Common target tracking algorithms include YOLOv3 algorithm (deep learning regression detection algorithm) and KCF algorithm (kernel-related filter tracking algorithm). For example, the Chinese patent application with publication number of CN112907553A discloses a method for detecting a high-definition image target based on YOLOv3, which applies the characteristic of high detection efficiency of the YOLOv3 algorithm and can realize rapid detection of the high-definition image target.
But in the face of the feature that petroleum drilling engineering operations are often exposed to complex operating scenarios, the YOLOv3 algorithm is no longer used. The problem of background interference caused by interference, illumination change and object shielding often exists in images acquired in complex operation scenes, and the YOLOv3 algorithm cannot provide higher accuracy, so that the target detection effect is poor.
Disclosure of Invention
The invention aims to provide a method for detecting illegal behaviors in a working scene, which is used for solving the problem of poor detection effect caused by utilizing a YOLOv3 algorithm model to detect the illegal behaviors.
In order to achieve the above object, the present invention provides a method for detecting illegal behaviors in a job scene, comprising the following steps:
1) Acquiring a rule-breaking behavior picture set in a working scene, wherein each picture in the rule-breaking behavior picture set is marked with various types of rule-breaking behaviors;
2) Training the constructed YOLOv3 algorithm model by taking the scene picture set of the field operation as a training set to obtain a violation behavior detection model; in training the YOLOv3 algorithm model, the loss function used includes a target positioning loss, which is:
where LOSS1 represents a LOSS of target positioning; lambda (lambda) coord Penalty coefficients representing coordinate predictions;a j-th bounding box representing an i-th cell is responsible for the target; s is S 2 To represent the number of units; b represents the number of predicted frames per unit; l (L) GIoU Representing a loss function between the predicted box and the real box; />A represents a prediction frame; a is that gt Representing a real frame; c represents a group comprising AAnd A gt Is the smallest frame of (2);
3) And acquiring a scene picture of the field operation, and detecting the illegal behaviors of operators in the scene of the field operation by using the illegal behavior detection model.
The beneficial effects are as follows: the invention improves the loss function of the YOLOv3 algorithm model and adopts L GIoU The loss function replaces the IOU loss function of the traditional YOLOv3 algorithm, namely, a minimum external frame is introduced on the basis of the IOU characteristics, so that the problem that the loss is equal to 0 when the detection frame and the real frame are not overlapped, and therefore, the target detection accuracy is improved, the recognition of smaller targets is facilitated, particularly, the background interference problem caused by interference, illumination change and object shielding is solved, the detection effect can be effectively improved, and the safety of drilling operation is guaranteed.
Further, in the training process of the YOLOv3 model, the used loss function further comprises a target confidence loss and a target category loss, wherein the target confidence loss and the target category loss are respectively:
where LOSS2 represents a target confidence LOSS;and->The true confidence of the jth bounding box of the ith unit where the target is located and the confidence of the prediction box are respectively represented; lambda (lambda) noobj Representing confidence penalty coefficients when the target is not included; />Indicating that the target has not fallenEntering the jth bounding box of the ith cell; LOSS3 represents target class LOSS; />And->Probability values of the real frame and the predicted frame category of the jth boundary frame of the ith unit where the target is located are respectively represented; class represents the number of all categories;
the LOSS function LOSS used accordingly is: loss=loss 1+loss2+loss3.
Further, each picture in the rule-breaking behavior picture set in the step 1) is a picture subjected to picture preprocessing.
The beneficial effects are as follows: each picture in the illegal activity picture set is a picture subjected to picture preprocessing, so that the prediction precision of the illegal activity detection model can be improved.
Further, the various types of violations include at least two of unworn/improperly worn work clothes, work shoes, helmets, safety belts, hydrogen sulfide detectors, and air respirators.
The beneficial effects are as follows: labeling various types of illegal behaviors ensures that the illegal behavior detection model can identify various illegal behaviors.
Further, the image preprocessing includes image preprocessing by using a CLAHE image enhancement processing algorithm, and the image preprocessing by using the CLAHE image enhancement processing algorithm includes the following steps: converting the color space of the input picture from RGB to HSV; performing block processing on the picture after the color space conversion, and performing histogram equalization operation on V components of each block in the HSV color space; and splicing the brightness V component of each block with the original S component and H component, and transferring to an RGB color space to realize the enhancement processing of the picture.
The beneficial effects are as follows: the CLAHE image enhancement preprocessing method is used for eliminating the influence of illumination change in a natural scene on the target recognition effect, and the accuracy of the trained illegal behavior detection model is ensured, so that the method is beneficial to being applied to drilling operation scenes in complex environments such as background interference, surrounding object shielding and illumination change.
Further, the size of each picture in the rule-breaking action picture set in step 1) is 416×416.
Further, in step 3), if the result of detecting the illegal behaviors is that certain type of illegal behaviors occur, corresponding alarm is performed.
The beneficial effects are as follows: alarming work is carried out when illegal behaviors are judged to timely remind operators, safety of the operators is guaranteed, and accidents are prevented.
Drawings
FIG. 1 is a flow chart diagram of a method for detecting offence in a job scenario of the present invention;
fig. 2 is a schematic diagram of structural connection of a system for implementing the method for detecting the behavior of violations in the operation scene of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent.
An embodiment of a method for detecting illegal behaviors in a job scene comprises the following steps:
in order to realize the method for detecting the illegal behaviors in the operation scene, the adopted system is shown in fig. 2 and comprises a camera module, a central processing module and an alarm feedback module. The camera module is a plurality of camera monitoring devices arranged in the field operation scene and is used for shooting the field operation scene, acquiring field operation video information and sending the shot image information to the central processing module. The central processing module is a processing unit with a data processing function, and the main function of the central processing module is to determine whether the illegal behaviors occur on site according to the data acquired by the camera module and combined with the constructed illegal behavior detection model. The alarm feedback module can be a device which can carry out sound alarm such as a broadcast loudspeaker and an interphone, and has the main function of carrying out alarm operation when the illegal behaviors are identified so as to remind operators, and other types of alarm devices can be arranged.
The invention relates to a method for detecting illegal behaviors in a working scene, the whole flow of which is shown in figure 1, comprising the following steps:
step one, constructing a rule-breaking behavior picture set. And collecting a large number of offence video materials through a plurality of camera monitoring devices, screening the offence video materials, and cutting the screened offence video materials into offence images with the size of 416 multiplied by 416 one to obtain an offence picture set.
And step two, preprocessing the illegal action picture. And (3) extracting the rule-breaking behavior picture set obtained in the step one, and preprocessing the rule-breaking behavior picture set by utilizing a CLAHE image enhancement preprocessing algorithm to obtain a sample set. The specific CLAHE image enhancement preprocessing algorithm comprises the following steps:
1) Converting the color space of the image from RGB to HSV;
2) Dividing the image into blocks, and performing histogram equalization operation on V (brightness) components of each block in HSV color space, wherein the operation performs shearing and average division on gray level pixels exceeding a threshold value in the histogram to each gray level;
3) Finally, the V component and the original H, S component are spliced and then transferred to an RGB color space to obtain an enhanced image.
And thirdly, labeling a sample set for training and testing. And (3) manually marking the sample set obtained after the rule-breaking behavior picture preprocessing, and dividing the marked sample set into a 70% training set and a 30% testing set. The labels herein are primarily labels that identify different violations, including but not limited to, not wearing or not properly wearing work clothes, work shoes, helmets, safety belts, hydrogen sulfide detectors, and air respirators.
And step four, constructing a violation behavior detection model. And constructing a YOLOv3 algorithm model, and learning and training the training set of the third 70% based on the YOLOv3 algorithm model to form the violation behavior detection model. Compared with the YOLOv3 algorithm model in the prior art, the structure of the YOLOv3 algorithm model used in the method does not occur, only the used loss functions are improved, and the used loss functions comprise target positioning loss, target confidence loss and target category loss, and specifically are as follows:
LOSS＝LOSS1+LOSS2+LOSS3
where LOSS represents the LOSS function used in model training; LOSS1 represents a LOSS of target localization; lambda (lambda) coord Penalty coefficients representing coordinate predictions;indicating whether the jth bounding box of the ith cell is responsible for this target object, if so +.>1, otherwise 0; s is S 2 To represent the number of units; b represents the number of predicted frames per unit; l (L) GIoU Representing a loss function between the predicted box and the real box; />A represents a prediction frame; a is that gt Representing a real frame; c represents a compound comprising A and A gt Is the smallest frame of (2); LOSS2 represents target confidence LOSS; />And->True confidence and pre-prediction of the jth bounding box of the ith cell in which the target is located, respectivelyMeasuring the confidence level of the frame; lambda (lambda) noobj Representing confidence penalty coefficients when the target is not included; />A j-th bounding box indicating that the object does not fall into the i-th cell; LOSS3 represents target class LOSS; />And->Probability values of the real frame and the predicted frame category of the jth boundary frame of the ith unit where the target is located are respectively represented; class represents the number of all categories.
And fifthly, testing a violation behavior detection model. Extracting 30% of the test set in the third step, and verifying the effect of the violation behavior detection model in the fourth step by using the test set; if the accuracy of the verification result reaches a threshold (which can be set to 95%), the model is output.
And step six, detecting a target. And acquiring a scene picture of the field operation, detecting the illegal behaviors of operators in the scene of the field operation by using the illegal behavior detection model, and alarming by using an alarm feedback module when the illegal behaviors are judged to occur.
In summary, the invention has the following characteristics: 1) The invention adopts L GIoU The loss function replaces the mean square error function of the traditional YOLOv3 algorithm, namely, a minimum external frame is introduced on the basis of the IOU characteristics, so that the problem that loss is equal to 0 and cannot be optimized when a detection frame and a real frame are not overlapped is solved, the target detection accuracy is improved, the identification of smaller targets is facilitated, and further the operation safety is guaranteed. 2) The CLAHE image enhancement preprocessing method is used for eliminating the influence of illumination change in a natural scene on the target recognition effect, so that the method is beneficial to being applied to drilling operation scenes in complex environments such as background interference, surrounding object shielding and illumination change.
The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art, who is within the scope of the present invention, should make equivalent substitutions or modifications according to the technical scheme of the present invention and the inventive concept thereof, and should be covered by the scope of the present invention.

Claims (7)

1. A method for detecting illegal behaviors in a job scene is characterized by comprising the following steps:
1) Acquiring a rule-breaking behavior picture set in a working scene, wherein each picture in the rule-breaking behavior picture set is marked with various types of rule-breaking behaviors;
2) Training the constructed YOLOv3 algorithm model by taking the scene picture set of the field operation as a training set to obtain a violation behavior detection model; in training the YOLOv3 algorithm model, the loss function used includes a target positioning loss, which is:
where LOSS1 represents a LOSS of target positioning; lambda (lambda) coord Penalty coefficients representing coordinate predictions;a j-th bounding box representing an i-th cell is responsible for the target; s is S 2 To represent the number of units; b represents the number of predicted frames per unit; l (L) GIoU Representing a loss function between the predicted box and the real box; />A represents a prediction frame; a is that gt Representing a real frame; c represents a compound comprising A and A gt Is the smallest frame of (2);
3) And acquiring a scene picture of the field operation, and detecting the illegal behaviors of operators in the scene of the field operation by using the illegal behavior detection model.
2. The method for detecting the behavior of violations in a job scenario according to claim 1, wherein in training the YOLOv3 model, the loss function used further includes a target confidence loss and a target class loss, where the target confidence loss and the target class loss are respectively:
where LOSS2 represents a target confidence LOSS;and->The true confidence of the jth bounding box of the ith unit where the target is located and the confidence of the prediction box are respectively represented; lambda (lambda) noobj Representing confidence penalty coefficients when the target is not included; />A j-th bounding box indicating that the object does not fall into the i-th cell; LOSS3 represents target class LOSS; p (P) i j And->Probability values of the real frame and the predicted frame category of the jth boundary frame of the ith unit where the target is located are respectively represented; class represents the number of all categories;
the LOSS function LOSS used accordingly is: loss=loss 1+loss2+loss3.
3. The method for detecting the behavior of violations in a job scenario according to claim 1, wherein each picture in the picture set of the behavior of violations in step 1) is a picture subjected to picture preprocessing.
4. The method of claim 1, wherein the various types of violations include at least two of unworn/unworn work wear, work shoes, helmets, safety belts, hydrogen sulfide detectors, and air respirators.
5. A method for detecting a violation in a job scene according to claim 3, wherein the picture preprocessing includes picture preprocessing by using a CLAHE picture enhancement processing algorithm, and the picture preprocessing by using the CLAHE picture enhancement processing algorithm includes the steps of:
converting the color space of the input picture from RGB to HSV;
performing block processing on the picture after the color space conversion, and performing histogram equalization operation on V components of each block in the HSV color space;
and splicing the brightness V component of each block with the original S component and H component, and transferring to an RGB color space to realize the enhancement processing of the picture.
6. The method for detecting the behavior of violations in a job scenario according to claim 1, wherein the size of each picture in the picture set of the behavior of violations in step 1) is 416×416.
7. The method for detecting the illegal behaviors in the operation scene according to claim 1, wherein in the step 3), if the result of detecting the illegal behaviors is that a certain type of illegal behaviors occurs, a corresponding alarm is given.
CN202210938993.1A 2022-08-05 2022-08-05 Method for detecting illegal behaviors in operation scene Pending CN117576719A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210938993.1A CN117576719A (en) 2022-08-05 2022-08-05 Method for detecting illegal behaviors in operation scene

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210938993.1A CN117576719A (en) 2022-08-05 2022-08-05 Method for detecting illegal behaviors in operation scene

Publications (1)

Publication Number Publication Date
CN117576719A true CN117576719A (en) 2024-02-20

Family

ID=89886838

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210938993.1A Pending CN117576719A (en) 2022-08-05 2022-08-05 Method for detecting illegal behaviors in operation scene

Country Status (1)

Country Link
CN (1) CN117576719A (en)

Similar Documents

Publication Publication Date Title
CN111898514B (en) Multi-target visual supervision method based on target detection and action recognition
CN110084165B (en) Intelligent identification and early warning method for abnormal events in open scene of power field based on edge calculation
CN111881730A (en) Wearing detection method for on-site safety helmet of thermal power plant
CN112183313B (en) SlowFast-based power operation field action identification method
CN113516076A (en) Improved lightweight YOLO v4 safety protection detection method based on attention mechanism
CN113903081A (en) Visual identification artificial intelligence alarm method and device for images of hydraulic power plant
CN113283344A (en) Mining conveying belt deviation detection method based on semantic segmentation network
CN112184773A (en) Helmet wearing detection method and system based on deep learning
CN112163572A (en) Method and device for identifying object
CN115035088A (en) Helmet wearing detection method based on yolov5 and posture estimation
CN112541393A (en) Transformer substation personnel detection method and device based on deep learning
CN113111771A (en) Method for identifying unsafe behaviors of power plant workers
CN112800975A (en) Behavior identification method in security check channel based on image processing
CN113298077A (en) Transformer substation foreign matter identification and positioning method and device based on deep learning
CN116310922A (en) Petrochemical plant area monitoring video risk identification method, system, electronic equipment and storage medium
CN114067268A (en) Method and device for detecting safety helmet and identifying identity of electric power operation site
CN113569956A (en) Mountain fire disaster investigation and identification method based on AI algorithm
CN117475353A (en) Video-based abnormal smoke identification method and system
CN116682162A (en) Robot detection algorithm based on real-time video stream
CN112488213A (en) Fire picture classification method based on multi-scale feature learning network
CN117423157A (en) Mine abnormal video action understanding method combining migration learning and regional invasion
CN117035669A (en) Enterprise safety production management method and system based on artificial intelligence
CN115171006B (en) Detection method for automatically identifying person entering electric power dangerous area based on deep learning
CN117576719A (en) Method for detecting illegal behaviors in operation scene
CN115829324A (en) Personnel safety risk silent monitoring method

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