CN113537019A - Detection method for identifying wearing of safety helmet of transformer substation personnel based on key points - Google Patents
Detection method for identifying wearing of safety helmet of transformer substation personnel based on key points Download PDFInfo
- Publication number
- CN113537019A CN113537019A CN202110768360.6A CN202110768360A CN113537019A CN 113537019 A CN113537019 A CN 113537019A CN 202110768360 A CN202110768360 A CN 202110768360A CN 113537019 A CN113537019 A CN 113537019A
- Authority
- CN
- China
- Prior art keywords
- wearing
- safety helmet
- detection
- image
- person
- 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
Links
Images
Landscapes
- Image Analysis (AREA)
Abstract
A detection method for recognizing the wearing of safety helmets of substation personnel based on key points comprises the following steps: firstly, establishing a safety helmet wearing label data set of substation personnel, and training the label data set; then, identifying key points of the human body by utilizing the images of the personnel wearing the helmet to be detected in the trained AlphaPose model; and finally, triangulating the generated human body key points to determine the head position, and detecting the wearing of the safety helmet by adopting a trained SSD network target detection model. Compared with other patent documents, the method can calculate according to the position information of the nose, the left shoulder and the right shoulder to obtain the detection area, and is more targeted and more accurate. Meanwhile, the calculation speed is obviously improved.
Description
Technical Field
The invention discloses a detection method for identifying wearing of safety helmets of transformer substation personnel based on key points, and belongs to the technical field of safety detection of transformer substation personnel.
Background
At present, relevant departments put high requirements on safe construction. When the transformer substation works, each constructor must wear a safety helmet, and the damage of dangerous accidents to the constructors and the loss of life and property can be reduced by wearing the safety helmet. Because human eye detection can not guarantee that each constructor wears the safety helmet all the time, at present, all install supervisory equipment in the transformer substation, whether utilize deep learning method and computer vision technique automated inspection constructor to wear the safety helmet and provide feedback, but current safety helmet wears detection algorithm and has the detection speed slow, detection precision not high scheduling problem, and when constructor is in complicated gesture such as bowing, crouching down or lying back, correctly discerns the safety helmet more difficult.
In view of this, the present technical field discloses the following patent documents:
chinese patent document CN111414825A discloses a helmet wearing detection method, which relates to image processing and recognition, and mainly solves the technical problems of poor robustness and poor adaptability existing in the current visual detection of helmet wearing, and the detection method comprises the following steps: acquiring an original image of a construction site, and amplifying the original image by N times through image transformation to obtain an amplified image; labeling safety helmet information on the amplified image; inputting the original image, the amplified image and the safety helmet information into a target detection network to learn and extracting image features to obtain human face features; obtaining a human face block diagram according to the human face features; and determining a head area according to the human face block diagram, and detecting a safety helmet in the head area. The document mainly identifies the human face part, and does not perform relevant correlation processing on the posture action of a constructor.
Chinese patent document CN110414400A discloses an automatic detection method and system for wearing a safety helmet on a construction site, which includes: acquiring a construction site monitoring video in real time, detecting all faces in the monitoring video, and acquiring a boundary frame of each face; tracking a boundary frame of a subsequent frame of face in the monitoring video based on a face tracking algorithm, executing a face key point detection algorithm, detecting the obtained face key points, and correcting a rectangular frame of the face tracking algorithm by using the face boundary frame of the current face key point detection result; obtaining a face Euler angle based on the detection result of the face key point and a PnP algorithm; detecting and tracking a face, taking a safety helmet area with the best Euler angle obtained by the face as a candidate area of the safety helmet, and judging whether the safety helmet exists in the candidate area of the safety helmet by using a safety helmet identification algorithm. The document combines near-face detection and a safety helmet detection and recognition system, so as to improve reliability for the recognition of a safety helmet, but the document mainly recognizes and tracks key points of a face, but when workers in different postures cannot show the face part, whether the detection effect can be maintained at a high level or not is difficult to predict.
Chinese patent document CN110287787A discloses an image recognition method, which includes: acquiring a working area video acquired by monitoring equipment; extracting an image containing a user from the working area video, wherein the image contains a human body target area of the user; determining a human head region in the human target region according to the head-shoulder feature matching template; extracting the directional gradient histogram characteristics and the color characteristics of the human head region; inputting the directional gradient histogram features into a first support vector machine model for human body classification, and determining whether the identified human head region contains a head target; if yes, inputting the color features into a second support vector machine model for color classification, and determining the color of the head target; and if the color of the head target is not the preset hat color, determining that the user does not wear the hat. This document mainly determines whether a worker is wearing a safety helmet by recognizing the color of a head target, and the use of color-defined helmets has certain technical limitations.
Chinese patent document CN111476083A provides an automatic identification method for wearing of safety helmets of power employees, which relates to the technical field of mode identification and intelligent video analysis, establishes a pedestrian detection model taking a pedestrian training sample as input, and realizes pedestrian area detection of a test sample through network parameter fine adjustment; the method comprises the steps of establishing an SSD (solid State disk) helmet detection model taking a helmet training sample set as input, inputting images of the upper half body area of pedestrians obtained by the pedestrian detection model into the trained SSD helmet detection model, and achieving real-time and high-accuracy automatic helmet wearing identification through SSD network parameter fine adjustment.
Chinese patent document CN110135290A provides a method and a system for detecting wearing of a safety helmet based on SSD and AlphaPose, the method comprising: firstly, identifying people and safety helmets in the image by using an SSD algorithm; then, estimating the human body posture of each person in the image by adopting an AlphaPose algorithm to obtain the pixel coordinates of the important joint points of each person; and finally, judging each person respectively to identify the person who does not wear the safety helmet. According to the technical scheme provided by the document, through the fusion of two models, namely SSD and AlphaPose, whether a human body wears a safety helmet or not can be effectively identified under various postures of the human body, however, the document judges each person respectively according to the pixel coordinates of important joint points of each person and the pixel coordinates of any two opposite angle vertexes of a minimum rectangular surrounding frame of each safety helmet so as to identify all people who do not wear the safety helmet, and the algorithm of the judging mode is complex.
In summary, how to provide an intelligent identification method for wearing a safety helmet with high efficiency and high precision, which is to efficiently identify the wearing condition of the safety helmet in a transformer substation environment, accurately identify the wearing condition of the safety helmet for constructors in complex postures, reduce potential safety hazards in the operation process of the transformer substation, and protect the personal safety of the constructors is a technical problem to be solved urgently by technical personnel in the field at present.
Disclosure of Invention
Aiming at the defects of the prior art, the invention discloses a detection method for identifying the wearing of safety helmets of transformer substation personnel based on key points. The invention can realize real-time detection and identification of the wearing condition of the safety helmet of the transformer substation personnel, and timely feed back, thereby reducing the safety accident rate caused by the fact that the constructor does not wear the safety helmet. After the method is implemented, the detection accuracy and real-time performance of the method meet the actual requirements of safety helmet detection of personnel in a transformer substation.
Summary of the invention:
a detection method for recognizing the wearing of safety helmets of substation personnel based on key points comprises the following steps: firstly, establishing a safety helmet wearing label data set of substation personnel, and training the label data set; then, identifying key points (a nose, a left shoulder and a right shoulder) of a human body in the image of the person wearing the helmet to be detected by utilizing the trained AlphaPose model; and finally, triangulating the generated human body key points to determine the head position, and detecting the wearing of the safety helmet by adopting a trained SSD network target detection model. Compared with other patent documents, the method can calculate according to the position information of the nose, the left shoulder and the right shoulder to obtain the detection area (namely the head area), and is more targeted and more accurate. Meanwhile, the calculation speed is obviously improved.
The detailed technical scheme of the invention is as follows:
a detection method for recognizing wearing of safety helmets of substation personnel based on key points is characterized by comprising the following steps:
1) establishing a safety helmet wearing label data set of transformer substation personnel, and sending the label data set into a target detection model: an SSD network target detection model;
2) identifying key points of a human body by using a trained AlphaPose model according to images of a person wearing the safety helmet to be detected;
3) performing triangulation positioning on the generated human body key points to determine the head position, and performing safety helmet wearing detection by adopting a trained SSD network target detection model;
wherein the method for determining the head position by triangulation comprises the following steps:
3-1) obtaining the position information of the key part according to the key points of the human body, namely calculating according to the position information of the nose, the left shoulder and the right shoulder to obtain the detected head area, and specifically comprising the following steps:
(3-1-1) determining a triangular area through the positions of the nose, the left shoulder and the right shoulder; the positions of the nose, the left shoulder and the right shoulder are respectively:
H(x0,y0),Ls(x1,y1),Rs(x2,y2) (I)
in formula (I), x0，y0The horizontal and vertical coordinates in the image of the nose are shown; x is the number of1,y1The horizontal and vertical coordinates in the image of the left shoulder; x is the number of2,y2The horizontal and vertical coordinates in the image of the right shoulder;
(3-1-2) rotating the triangular area by 180 degrees with the nose as the center to obtain a new triangle, and setting the positions of the left shoulder and the right shoulder after rotation as follows:
Ls′(x1′,y1′),Rs′(x2′,y2′) (II)
in formula (II), x1′,y1' is the horizontal and vertical coordinates in the image where the left shoulder is rotated; x is the number of2′,y2' is the horizontal and vertical coordinates in the image where the right shoulder is rotated;
the positions of the left shoulder and the right shoulder after rotation are as follows:
(Ls′(x1′,y1′),Rs′(x2′,y2′))＝(Ls′(2x0-x1，2y0-y1),Rs′(2x0-x2,2y0-y2)) (III)
(3-1-3) forming a quadrangle together according to the original triangle obtained in the step (3-1-1) and the new triangle obtained by rotating in the step (3-1-2), wherein the lengths of diagonal lines forming the quadrangle are respectively as follows:
(3-1-4) taking the longest diagonal line of the quadrangle in the step (3-1-3) as the diameter and the intersection point of the diagonal lines as the circle center to make the maximum excircle, and then obtaining the circumscribed square of the excircle, wherein the area framed and selected by the circumscribed square is the head area in the step (3-1);
3-2) detecting the image of the person wearing the safety helmet to be detected, comprising the following steps:
(3-2-1) selecting a detection area, namely a head area, from the step (3-1-4), and carrying out safety helmet wearing detection by using the trained AlphaPose model as a target detection model;
(3-2-2) obtaining a detection result of the wearing state of the safety helmet according to the step (3-2-1):
if the person wears the safety helmet, the detection is continued;
if the person does not wear the safety helmet, a feedback result is given. And returning the feedback result of the person who does not wear the safety helmet to the platform in real time for warning.
Preferably according to the present invention, the feedback result includes but is not limited to: plain text information of the person not wearing the safety helmet or a frame selection identification of the person not wearing the safety helmet. The warning is not limited to the image of the framed logo, and may also include plain text information, that is, the information form of the behavior occurrence region that only identifies the person who does not wear the safety helmet belongs to the technical content to be protected by the present invention.
Preferably, the step 1) of establishing a helmet wearing annotation data set of the substation personnel comprises the following steps:
1-1) obtaining a picture of a transformer substation worker in a transformer substation video stream intercepting manner;
1-2) acquiring images of single person and multiple persons in various postures corresponding to the pictures in the step 1-1) in various operation places;
1-3) amplifying the image in the step 1-2) to obtain an image data set to be annotated.
Preferably, in step 1), the step of establishing a helmet wearing annotation data set of the substation personnel further includes:
1-4) after the image data set is obtained according to the step 1-3), marking the transformer substation personnel in each picture to obtain a marked image data set:
labeling a target area and a type by using labelimg for the labeling of the sample, and labeling a transformer substation worker wearing a safety helmet with a helmet, namely a positive sample; substation personnel without a safety helmet are marked head, i.e. a negative sample.
Preferably, in step 3), the training of the SSD network target detection model includes the following steps:
3-3) introducing a characteristic pyramid network structure, FPN, into the classical SSD network; the optimized SSD network structure enables the feature map finally used for prediction to have semantic information of shallow features and deep features at the same time, and has strong adaptability to various detection environments, so that the robustness of wearing detection of the safety helmet is improved; the classic SSD network is a classic one-stage algorithm, has a faster detection rate compared with a two-stage method, but has a poor detection effect robustness under the interference of complex environmental factors; the structure diagrams of the SSD network and the SSD network with the improved structure by introducing the characteristic pyramid network are respectively shown in the attached figures 1 and 2;
3-4) adopting the SSD model optimized in the step 3-3) as a target detection model, and training the marked image data set, wherein the adopted feature extraction network is VGG-16.
According to the invention, the image of the person wearing the helmet to be detected is acquired by the following method:
the method comprises the steps that video monitoring equipment erected in a transformer substation is utilized to obtain transformer substation video information in real time to a server, and monitoring images are obtained in real time in a transformer substation video stream intercepting mode.
Preferably, the specific steps of step 2) include:
2-1) selecting an AlphaPose model containing 26 key point models, wherein the adopted feature extraction network is resnet-50;
2-2) if no person exists in the image of the person wearing the helmet to be detected, no detection is carried out; and if the image of the person wearing the helmet to be detected contains the person, identifying the key points to obtain 26 key points of the human body of the person.
The technical advantages of the invention are as follows:
1) the invention provides an optimized SSD algorithm, introduces a characteristic pyramid network structure, and improves the adaptability of the network to various detection environments aiming at the complexity and the particularity of a construction operation scene, thereby improving the robustness of the wearing detection of the safety helmet.
2) The invention provides a human body key point detection algorithm, solves the problem that the relative position of a safety helmet and a human body under the complex posture of a constructor is difficult to determine, and improves the recognition rate and the accuracy rate of safety helmet detection.
3) Aiming at the high difficulty of the wearing detection of the safety helmet in a complex posture, the invention provides a key point detection algorithm to determine the head area according to a triangulation method, then adopts an optimized first-order target detection network to solve the problem of polar imbalance between the safety helmet and a complex construction scene, trains a model to detect the wearing condition of the safety helmet, finally adopts the detection method of the invention to achieve the omission factor of 8.21%, the false detection rate of 6.98% and the precision of 95.25%, and meets the actual requirements of a transformer substation on the wearing detection of the safety helmet.
Drawings
FIG. 1 is a block diagram of a classic SSD network of the present invention;
FIG. 2 is a diagram of an optimized SSD network architecture according to the present invention;
FIG. 3 is a flow chart of the identification method of the present invention;
FIG. 4 is an annotated image in an embodiment of the invention;
FIG. 5 is a monitoring image acquired in real time in an embodiment of the present invention;
FIG. 6 is an image of FIG. 5 after the head region has been located in an embodiment of the present invention;
FIG. 7 is an image of FIG. 5 after detection and labeling of the feedback result according to an embodiment of the present invention.
Detailed Description
The present invention is further illustrated by, but not limited to, the following examples.
Examples 1,
The image of the person wearing the helmet to be detected is acquired by the following method: the method comprises the steps that video monitoring equipment erected in a transformer substation is utilized to obtain transformer substation video information in real time to a server, and monitoring images are obtained in real time in a transformer substation video stream intercepting mode.
As shown in fig. 3, a detection method for identifying wearing of safety helmets of substation personnel based on key points includes:
1) establishing a safety helmet wearing label data set of transformer substation personnel, and sending the label data set into a target detection model: an SSD network target detection model;
2) identifying key points of a human body by using a trained AlphaPose model according to images of a person wearing the safety helmet to be detected; the specific steps of step 2) include:
2-1) selecting an AlphaPose model containing 26 key point models, wherein the adopted feature extraction network is resnet-50;
2-2) if no person exists in the image of the person wearing the helmet to be detected, no detection is carried out; if the image of the person wearing the helmet to be detected contains the person, 26 key points of the human body of the person are obtained after key point identification;
3) performing triangulation positioning on the generated human body key points to determine the head position, and performing safety helmet wearing detection by adopting a trained SSD network target detection model;
wherein the method for determining the head position by triangulation comprises the following steps:
3-1) obtaining the position information of the key part according to the key points of the human body, namely calculating according to the position information of the nose, the left shoulder and the right shoulder to obtain the detected head area, and specifically comprising the following steps:
(3-1-1) determining a triangular area through the positions of the nose, the left shoulder and the right shoulder; the positions of the nose, the left shoulder and the right shoulder are respectively:
H(x0,y0),Ls(x1,y1),Rs(x2,y2) (I)
in formula (I), x0，y0The horizontal and vertical coordinates in the image of the nose are shown; x is the number of1,y1The horizontal and vertical coordinates in the image of the left shoulder; x is the number of2,y2The horizontal and vertical coordinates in the image of the right shoulder;
(3-1-2) rotating the triangular area by 180 degrees with the nose as the center to obtain a new triangle, and setting the positions of the left shoulder and the right shoulder after rotation as follows:
Ls′(x1′,y1′),Rs′(x2′,y2′) (II)
in formula (II), x1′,y1' is the image where the left shoulder is rotatedThe inside horizontal and vertical coordinates; x is the number of2′,y2' is the horizontal and vertical coordinates in the image where the right shoulder is rotated;
the positions of the left shoulder and the right shoulder after rotation are as follows:
(Ls′(x1′,y1′),Rs′(x2′,y2′))＝(Ls′(2x0-x1，2y0-y1),Rs′(2x0-x2,2y0-y2)) (III)
(3-1-3) forming a quadrangle together according to the original triangle obtained in the step (3-1-1) and the new triangle obtained by rotating in the step (3-1-2), wherein the lengths of diagonal lines forming the quadrangle are respectively as follows:
(3-1-4) taking the longest diagonal line of the quadrangle in the step (3-1-3) as the diameter and the intersection point of the diagonal lines as the circle center to make the maximum excircle, and then obtaining the circumscribed square of the excircle, wherein the area framed and selected by the circumscribed square is the head area in the step (3-1);
3-2) detecting the image of the person wearing the safety helmet to be detected, comprising the following steps:
(3-2-1) selecting a detection area, namely a head area, from the step (3-1-4), and carrying out safety helmet wearing detection by using the trained AlphaPose model as a target detection model;
(3-2-2) obtaining a detection result of the wearing state of the safety helmet according to the step (3-2-1):
if the person wears the safety helmet, the detection is continued;
if the person does not wear the safety helmet, a feedback result is given. And returning the feedback result of the person who does not wear the safety helmet to the platform in real time for warning.
The feedback results include, but are not limited to: plain text information that the person did not wear a hard hat, or a frame selection identification that the person did not wear a hard hat. The warning is not limited to the image of the framed logo, and may also include plain text information, that is, the information form of the behavior occurrence region that only identifies the person who does not wear the safety helmet belongs to the technical content to be protected by the present invention.
As shown in fig. 4, the step 1) of establishing a helmet wearing annotation data set of substation personnel includes:
1-1) obtaining a picture of a transformer substation worker in a transformer substation video stream intercepting manner;
1-2) acquiring images of single person and multiple persons in various postures corresponding to the pictures in the step 1-1) in various operation places;
1-3) amplifying the image in the step 1-2) to obtain an image data set to be labeled;
1-4) after the image data set is obtained according to the step 1-3), marking the transformer substation personnel in each picture to obtain a marked image data set:
labeling a target area and a type by using labelimg for the labeling of the sample, and labeling a transformer substation worker wearing a safety helmet with a helmet, namely a positive sample; substation personnel without a safety helmet are marked head, i.e. a negative sample.
Examples 2,
As shown in fig. 1 and 2, in the detection method for identifying the wearing of the safety helmet of the substation personnel based on the key point according to embodiment 1, the step 3) of training the SSD network target detection model includes the following steps:
3-3) introducing a characteristic pyramid network structure, FPN, into the classical SSD network;
3-4) adopting the SSD model optimized in the step 3-3) as a target detection model, and training the marked image data set, wherein the adopted feature extraction network is VGG-16.
Application examples,
The following describes in detail a practical application scenario with reference to embodiments 1 and 2.
A detection method for identifying the wearing of safety helmets of transformer substation personnel based on key points comprises the following steps:
s1: acquiring images of various postures of personnel on certain monitoring equipment of a certain transformer substation in A city of Shandong province to form an image data set to be marked, as shown in FIG. 4;
s2: carrying out wearing marking on the acquired image to be marked on the transformer substation personnel and the safety helmet thereof;
s3: sending the marked data set into an improved SSD network for training;
s4: acquiring real-time images of personnel through substation monitoring, as shown in fig. 5;
s5: putting the acquired image into a trained AlphaPose model for key point identification;
s6: triangulating the generated human body key points to determine the head position, as shown in fig. 6;
s7: and (3) adopting a trained target detection model to carry out safety helmet wearing detection on the determined head position, and detecting whether a safety helmet is worn or not, as shown in fig. 7.
In step S1, the step of acquiring the image data set to be labeled includes:
s11: acquiring substation video information in real time to a server by using video monitoring equipment erected in each substation, and acquiring pictures of constructors in a substation video stream intercepting manner;
s12: the method comprises the steps of collecting images of single people and multiple people in various postures in each operation place, simply screening, and selecting clear 8000 original images, wherein the collected images are the wearing conditions of the personnel in a construction site;
s13: carrying out operations such as turning, translation and the like on the data in the step S12 to carry out data augmentation, and obtaining 30000 image data sets to be annotated;
wherein, the step S2 includes:
s21: after the image data set is obtained according to step S13, the constructor in each image is labeled to obtain a labeled image data set, and the labeled image is as shown in fig. 4. Labeling of samples target areas and categories were labeled using labellimg, with the person wearing the helmet labeled helmet (positive sample) and the person not wearing the helmet labeled head (negative sample).
The step of training the optimized SSD model in step S3 includes:
s31: as shown in fig. 1 and 2, a classic SSD model is optimized, and a feature pyramid network structure (FPN) is introduced, which enables a feature map finally used for prediction to have semantic information of both shallow features and deep features.
S32: the method adopts the SSD model optimized in S31 as a target detection model, trains the target detection model through the data set obtained in S21, and adopts a feature extraction network of VGG-16.
In step S4, the manner of acquiring the real-time monitoring image is as follows:
s41: the method comprises the steps of utilizing video monitoring equipment erected in a corridor of a certain transformer substation in the Shandong province B city to obtain transformer substation video information in real time to a server, and obtaining a monitoring image in real time in a transformer substation video stream intercepting mode, wherein the monitoring image is shown in figure 5.
The human body key point identification method in step S5 is as follows:
s51: putting the images acquired in real time into a trained AlphaPose model for carrying out key point identification, wherein 26 key point models are adopted, and a feature extraction network adopted by the models is resnet-50;
and S52, if no person exists in the S51 image, no detection is carried out, and if the person exists, 26 key points of the human body of the constructor can be obtained after key point identification.
The triangulation manner in step S6 is as follows:
s61: the key points of the human body are obtained according to S52, the position information of the key parts is obtained, and for the current research, only three points (nose, left shoulder and right shoulder) are needed, and the detection region (head region) is obtained by calculating according to the position information of the three points of the nose, left shoulder and right shoulder, as shown in fig. 6.
S62: first, a triangular area is defined by three points of the nose, the left shoulder and the right shoulder, as shown in fig. 6. Wherein, the positions of the head, the left shoulder and the right shoulder are:
H(x0,y0),Ls(x1,y1),Rs(x2,y2)
s63: then, the triangular region is rotated by 180 degrees with the nose as the center to obtain a new triangle, and as shown in fig. 6, the positions of the left shoulder and the right shoulder after rotation are set as
Ls′(x1′,y1′),Rs′(x2′,y2′)
I.e. the positions of the left and right shoulders after rotation
(Ls′(x1′,y1′),Rs′(x2′,y2′))＝(Ls′(2x0-x1，2y0-y1),Rs′(2x0-x2,2y0-y2))
S64: as shown in fig. 6, the original triangle obtained in step S62 and the new triangle rotated in step S63 together form a quadrangle, and the length of the diagonal line of the quadrangle is
S65: then, as shown in fig. 6, the longest diagonal line of the quadrangle of step S64 is taken as the diameter, and the intersection point of the diagonal lines is taken as the center of the circle as the maximum outer circle. Then, a detection region (head region) which is a circumscribed square of the outer circle is obtained.
Wherein, the mode of the safety helmet wearing detection in the step S7 is as follows:
s71: and (4) performing safety helmet wearing detection on the detection area (head area) selected in the step (S65) by using the target detection model trained in the step (S32), and finishing the identification detection of safety helmet wearing of the transformer substation constructor, as shown in FIG. 7.
S72: obtaining a detection result of the wearing state of the safety helmet according to the step S71, and if the constructor wears the safety helmet, continuing the detection; and if the constructor does not wear the safety helmet, feedback is given, and the platform is returned to warn.
Claims (7)
1. A detection method for recognizing wearing of safety helmets of substation personnel based on key points is characterized by comprising the following steps:
1) establishing a safety helmet wearing label data set of transformer substation personnel, and sending the label data set into a target detection model: an SSD network target detection model;
2) identifying key points of a human body by using a trained AlphaPose model according to images of a person wearing the safety helmet to be detected;
3) performing triangulation positioning on the generated human body key points to determine the head position, and performing safety helmet wearing detection by adopting a trained SSD network target detection model;
wherein the method for determining the head position by triangulation comprises the following steps:
3-1) obtaining the position information of the key part according to the key points of the human body, namely calculating according to the position information of the nose, the left shoulder and the right shoulder to obtain the detected head area, and specifically comprising the following steps:
(3-1-1) determining a triangular area through the positions of the nose, the left shoulder and the right shoulder; the positions of the nose, the left shoulder and the right shoulder are respectively:
H(x0,y0),Ls(x1,y1),Rs(x2,y2) (I)
in formula (I), x0，y0The horizontal and vertical coordinates in the image of the nose are shown; x is the number of1,y1The horizontal and vertical coordinates in the image of the left shoulder; x is the number of2,y2The horizontal and vertical coordinates in the image of the right shoulder;
(3-1-2) rotating the triangular area by 180 degrees with the nose as the center to obtain a new triangle, and setting the positions of the left shoulder and the right shoulder after rotation as follows:
Ls′(x1′,y1′),Rs′(x2′,y2′) (II)
in formula (II), x1′,y1' is the horizontal and vertical coordinates in the image where the left shoulder is rotated; x is the number of2′,y2' is the horizontal and vertical coordinates in the image where the right shoulder is rotated;
the positions of the left shoulder and the right shoulder after rotation are as follows:
(Ls′(x1′,y1′),Rs′(x2′,y2′))＝(Ls′(2x0-x1，2y0-y1),Rs′(2x0-x2,2y0-y2)) (III)
(3-1-3) forming a quadrangle together according to the original triangle obtained in the step (3-1-1) and the new triangle obtained by rotating in the step (3-1-2), wherein the lengths of diagonal lines forming the quadrangle are respectively as follows:
(3-1-4) taking the longest diagonal line of the quadrangle in the step (3-1-3) as the diameter and the intersection point of the diagonal lines as the circle center to make the maximum excircle, and then obtaining the circumscribed square of the excircle, wherein the area framed and selected by the circumscribed square is the head area in the step (3-1);
3-2) detecting the image of the person wearing the safety helmet to be detected, comprising the following steps:
(3-2-1) selecting a detection area, namely a head area, from the step (3-1-4), and carrying out safety helmet wearing detection by using the trained AlphaPose model as a target detection model;
(3-2-2) obtaining a detection result of the wearing state of the safety helmet according to the step (3-2-1):
if the person wears the safety helmet, the detection is continued;
if the person does not wear the safety helmet, a feedback result is given.
2. The detection method for identifying the wearing of the safety helmet of the substation personnel based on the key points as claimed in claim 1, wherein the feedback result comprises but is not limited to: plain text information of the person not wearing the safety helmet or a frame selection identification of the person not wearing the safety helmet.
3. The detection method for identifying the wearing of the safety helmet of the substation personnel based on the key point as claimed in claim 1, wherein the step 1) of establishing the safety helmet wearing label data set of the substation personnel comprises the following steps:
1-1) obtaining a picture of a transformer substation worker in a transformer substation video stream intercepting manner;
1-2) acquiring images of single person and multiple persons in various postures corresponding to the pictures in the step 1-1) in various operation places;
1-3) amplifying the image in the step 1-2) to obtain an image data set to be annotated.
4. The detection method for identifying the wearing of the safety helmet of the substation personnel based on the key point as claimed in claim 3, wherein the step 1) of establishing the safety helmet wearing label data set of the substation personnel further comprises the following steps:
1-4) after the image data set is obtained according to the step 1-3), marking the transformer substation personnel in each picture to obtain a marked image data set:
labeling a target area and a type by using labelimg for the labeling of the sample, and labeling a transformer substation worker wearing a safety helmet with a helmet, namely a positive sample; substation personnel without a safety helmet are marked head, i.e. a negative sample.
5. The detection method based on key point identification of wearing of safety helmets of substation personnel according to claim 1, wherein the step 3) training of the SSD network target detection model comprises the following steps:
3-3) introducing a characteristic pyramid network structure, FPN, into the classical SSD network;
3-4) adopting the SSD model optimized in the step 3-3) as a target detection model, and training the marked image data set, wherein the adopted feature extraction network is VGG-16.
6. The detection method based on key point identification of wearing of safety helmets for substation personnel according to claim 1, wherein the images of personnel wearing the safety helmets to be detected are acquired by the following method:
the method comprises the steps that video monitoring equipment erected in a transformer substation is utilized to obtain transformer substation video information in real time to a server, and monitoring images are obtained in real time in a transformer substation video stream intercepting mode.
7. The detection method for identifying the wearing of the safety helmet of the substation personnel based on the key points as claimed in claim 1 is characterized in that the specific steps of the step 2) comprise:
2-1) selecting an AlphaPose model containing 26 key point models, wherein the adopted feature extraction network is resnet-50;
2-2) if no person exists in the image of the person wearing the helmet to be detected, no detection is carried out; and if the image of the person wearing the helmet to be detected contains the person, identifying the key points to obtain 26 key points of the human body of the person.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110768360.6A CN113537019A (en) | 2021-07-07 | 2021-07-07 | Detection method for identifying wearing of safety helmet of transformer substation personnel based on key points |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110768360.6A CN113537019A (en) | 2021-07-07 | 2021-07-07 | Detection method for identifying wearing of safety helmet of transformer substation personnel based on key points |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113537019A true CN113537019A (en) | 2021-10-22 |
Family
ID=78097956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110768360.6A Pending CN113537019A (en) | 2021-07-07 | 2021-07-07 | Detection method for identifying wearing of safety helmet of transformer substation personnel based on key points |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113537019A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116311082A (en) * | 2023-05-15 | 2023-06-23 | 广东电网有限责任公司湛江供电局 | Wearing detection method and system based on matching of key parts and images |
| CN117935341A (en) * | 2024-03-21 | 2024-04-26 | 福建信息职业技术学院 | Automatic sign-in method based on face recognition |
-
2021
- 2021-07-07 CN CN202110768360.6A patent/CN113537019A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116311082A (en) * | 2023-05-15 | 2023-06-23 | 广东电网有限责任公司湛江供电局 | Wearing detection method and system based on matching of key parts and images |
| CN117935341A (en) * | 2024-03-21 | 2024-04-26 | 福建信息职业技术学院 | Automatic sign-in method based on face recognition |
| CN117935341B (en) * | 2024-03-21 | 2024-06-04 | 福建信息职业技术学院 | Automatic sign-in method based on face recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Fang et al. | Detecting non-hardhat-use by a deep learning method from far-field surveillance videos | |
| CN109670441B (en) | Method, system, terminal and computer readable storage medium for realizing wearing recognition of safety helmet | |
| CN110502965B (en) | Construction safety helmet wearing monitoring method based on computer vision human body posture estimation | |
| Park et al. | Continuous localization of construction workers via integration of detection and tracking | |
| CN111460962B (en) | Face recognition method and face recognition system for mask | |
| CN111898514B (en) | Multi-target visual supervision method based on target detection and action recognition | |
| WO2020253308A1 (en) | Human-machine interaction behavior security monitoring and forewarning method for underground belt transportation-related personnel | |
| CN113516076B (en) | Attention mechanism improvement-based lightweight YOLO v4 safety protection detection method | |
| CN113743256B (en) | Intelligent early warning method and device for site safety | |
| CN110991315A (en) | Method for detecting wearing state of safety helmet in real time based on deep learning | |
| CN107977639A (en) | A kind of face definition judgment method | |
| CN113139437B (en) | Helmet wearing inspection method based on YOLOv3 algorithm | |
| CN113537019A (en) | Detection method for identifying wearing of safety helmet of transformer substation personnel based on key points | |
| Tan et al. | Dynamic hand gesture recognition using motion trajectories and key frames | |
| CN113903081A (en) | Visual identification artificial intelligence alarm method and device for images of hydraulic power plant | |
| CN109460704A (en) | A kind of fatigue detection method based on deep learning, system and computer equipment | |
| CN113011280A (en) | Method and device for detecting person contact distance, computer equipment and storage medium | |
| CN116259002A (en) | Human body dangerous behavior analysis method based on video | |
| Mei et al. | Human intrusion detection in static hazardous areas at construction sites: Deep learning–based method | |
| CN113807240A (en) | Intelligent transformer substation personnel dressing monitoring method based on uncooperative face recognition | |
| CN113111771A (en) | Method for identifying unsafe behaviors of power plant workers | |
| CN115223249A (en) | Quick analysis and identification method for unsafe behaviors of underground personnel based on machine vision | |
| Kwak et al. | A study on Detecting the Safety helmet wearing using YOLOv5-S model and transfer learning | |
| CN113076825A (en) | Transformer substation worker climbing safety monitoring method | |
| CN116311082B (en) | Wearing detection method and system based on matching of key parts and images |
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 |