CN109753898A - A kind of safety cap recognition methods and device - Google Patents
A kind of safety cap recognition methods and device Download PDFInfo
- Publication number
- CN109753898A CN109753898A CN201811570198.1A CN201811570198A CN109753898A CN 109753898 A CN109753898 A CN 109753898A CN 201811570198 A CN201811570198 A CN 201811570198A CN 109753898 A CN109753898 A CN 109753898A
- Authority
- CN
- China
- Prior art keywords
- network model
- neural network
- image
- convolution neural
- head zone
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of safety cap recognition methods and devices, for solving whether have safe wearing cap for staff, the not high problem of the accuracy of detection and identification.This method comprises: the image comprising human body to be identified is input in the first convolution neural network model that training is completed in advance, the fisrt feature figure of described image is obtained, wherein the fisrt feature figure includes the location information of head zone prediction block；According to the mapping relations of the location information of head zone prediction block in fisrt feature figure and the fisrt feature figure and described image, the subgraph in described image comprising head zone is determined；The subgraph is input in the second convolution neural network model that training is completed in advance, whether is obtained in the head zone for including in the subgraph comprising safety cap.
Description
Technical field
The present invention relates to deep learning and target identification technology field more particularly to a kind of safety cap recognition methods and dresses
It sets.
Background technique
In routine work and life, can all there are some security risks, asking for the inherently safe of the mankind may be endangered
Topic.It is the safety of safeguard work personnel such as when personnel ride motor on-site, in coal mine and on road,
Require safe wearing cap.Therefore at specific occasion, by the way that personnel, whether safe wearing cap is identified, can effectively subtract
The generation of few social danger.
Coal mine underground personnel positioning and tracking based on safety cap detection are disclosed in the prior art, and this method is to pacify
Based on full cap, by detecting to safety cap and tracking the positioning and tracking realized to coal mine down-hole personnel.It is practical in conjunction with coal mine,
The equal safe wearing cap of personnel in the pit shows to detect personnel targets if the presence of safety cap can be detected.Above-mentioned detection side
Method designs classifier using statistical pattern recognition method, safety cap detection is carried out with classifier, in conjunction with Kalman filter and Mean-
Shift tracks safety cap.
The above method is only applicable to the tracking of the personnel positioning under coal mine scene, is based on safety cap to coal mine down-hole personnel
Positioning and tracking are the detection of safety cap progress have been worn based on each personnel, and cannot judge whether staff has
Safe wearing cap, and this method is detected and is identified that the accuracy for detecting and identifying is not high using conventional method.
Summary of the invention
The embodiment of the invention provides a kind of safety cap recognition methods and devices, for whether safe wearing cap is examined
It surveys.
The embodiment of the invention provides a kind of safety cap recognition methods, comprising:
Image comprising human body to be identified is input in the first convolution neural network model that training is completed in advance, is obtained
The fisrt feature figure of described image, wherein the fisrt feature figure includes the location information of head zone prediction block；
According to the location information of head zone prediction block in fisrt feature figure and the fisrt feature figure and described image
Mapping relations, determine in described image include head zone subgraph；
The subgraph is input in the second convolution neural network model that training is completed in advance, obtains the subgraph
In include head zone in whether comprising safety cap.
Further, described that the image comprising human body to be identified is input to the first convolution nerve net that training is completed in advance
Before in network model, the method also includes:
Described image is zoomed into the first default size.
Further, described that the subgraph is input to it in the second convolution neural network model that training is completed in advance
Before, the method also includes:
The subgraph comprising head zone is zoomed into the second default size.
Further, the process for training the first convolution neural network model in advance includes:
Sample image is obtained, wherein being labelled with the location information of head zone frame in the sample image；
Each sample image is input in the first convolution neural network model, according to the first convolution neural network model
Each output is trained the first convolution neural network model.
Further, the process for training the second convolution neural network model in advance includes:
Sample image is obtained, wherein be labelled with the corresponding label information of the sample image in the sample image, wherein institute
State label information include wear a safety helmet label, wear common hat label, helmet label and label of not puing a label on；
Each sample image is input in the second convolution neural network model, according to the second convolution neural network model
Each output is trained the second convolution neural network model.
Further, the first convolution neural network model and the second convolution neural network model part are identical.
The embodiment of the invention provides a kind of safety cap identification device, which includes:
First obtains module, for will include that the image of human body to be identified is input to the first convolution mind that training is completed in advance
Through in network model, obtaining the fisrt feature figure of described image, wherein the fisrt feature figure includes head zone prediction block
Location information；
Determining module, for according to the location information of head zone prediction block in fisrt feature figure and first spy
The mapping relations of sign figure and described image determine the subgraph in described image comprising head zone；
Second obtains module, for the subgraph to be input to the second convolution neural network model that training is completed in advance
In, it whether obtains in the head zone for including in the subgraph comprising safety cap.
Further, described device further include:
First Zoom module, for described image to be zoomed to the first default size.
Further, described device further include:
Second Zoom module, for the subgraph comprising head zone to be zoomed to the second default size.
Further, described device further include:
First training module, for obtaining sample image, wherein being labelled with the position of head zone frame in the sample image
Confidence breath；Each sample image is input in the first convolution neural network model, according to the first convolution neural network model
Each output is trained the first convolution neural network model.
Further, described device further include:
Second training module, for obtaining sample image, wherein it is corresponding to be labelled with the sample image in the sample image
Label information, wherein the label information include wear a safety helmet label, wear common hat label, helmet and label and do not attach the names of pre-determined candidates
Subtab；Each sample image is input in the second convolution neural network model, according to the second convolution neural network model
Each output is trained the second convolution neural network model.
The embodiment of the present invention provides a kind of safety cap recognition methods and device, and this method is by the image comprising human body to be identified
It is input in the first convolution neural network model that training is completed in advance, obtains the fisrt feature figure of described image, wherein described
Fisrt feature figure includes the location information of head zone prediction block；According to the position letter of head zone prediction block in fisrt feature figure
The mapping relations of breath and the fisrt feature figure and described image determine the subgraph in described image comprising head zone；
The subgraph is input in the second convolution neural network model that training is completed in advance, obtains in the subgraph and include
It whether include safety cap in head zone.
Due in embodiments of the present invention, the image comprising human body to be identified being input to the first volume that training is completed in advance
In product neural network model, the fisrt feature figure of the image is obtained；According to the position of head zone prediction block in fisrt feature figure
The mapping relations of information and fisrt feature figure and the image comprising human body to be identified determine the subgraph comprising head zone；
Detected for the subgraph comprising head zone, can be determined more accurately the people whether safe wearing cap.To include
The subgraph of head zone is input in the second convolution neural network model that training is completed in advance, includes in acquisition subgraph
Whether include safety cap in head zone, in embodiments of the present invention, determines the subgraph comprising head zone in the picture first
Picture, identifies whether in the subgraph later comprising safety cap, so as to accurately judge people whether safe wearing cap, improve
The accuracy of identification.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly introduced, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill in field, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is a kind of flow diagram for safety cap recognition methods that the embodiment of the present invention 1 provides；
Fig. 2 is the first convolution neural network model that the embodiment of the present invention 3 provides and the second convolution neural network model
Structure chart；
Fig. 3 is a kind of idiographic flow schematic diagram for safety cap recognition methods that the embodiment of the present invention 6 provides；
Fig. 4 is a kind of safety cap identification device structural schematic diagram that the embodiment of the present invention 7 provides.
Specific embodiment
The present invention will be describe below in further detail with reference to the accompanying drawings, it is clear that described embodiment is only this
Invention a part of the embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art exist
All other embodiment obtained under the premise of creative work is not made, shall fall within the protection scope of the present invention.
Embodiment 1:
Fig. 1 is a kind of flow diagram of safety cap recognition methods provided in an embodiment of the present invention, which includes following
Step:
S101: the image comprising human body to be identified is input to the first convolution neural network model that training is completed in advance
In, obtain the fisrt feature figure of described image, wherein the fisrt feature figure includes the location information of head zone prediction block.
Since the major function of safety cap is to protect the head safety of the mankind, safety cap is only worn on head and could protect
The head safety of the mankind.Therefore before being identified to the safety cap in image, it is necessary first to including human body to be identified
Head zone in image is detected, and the subgraph comprising head zone is obtained.It include head according to what is obtained after detection operation
The subgraph in portion region carries out identification operation, determines whether people has safe wearing cap.
Image comprising human body to be identified is input in the first convolution neural network model, by the convolution sum of several layers
Pondization operates the fisrt feature figure for finally obtaining the image comprising human body to be identified, and in the fisrt feature figure includes head zone
The location information of prediction block.The location information of the head zone prediction block is used to determine the subgraph in image comprising head zone
Picture.Subsequent identification operation is carried out for the subgraph comprising head zone, obtained result is more accurate.
S102: according to the location information of head zone prediction block in fisrt feature figure and the fisrt feature figure and institute
The mapping relations for stating image determine the subgraph in described image comprising head zone.
The location information for the head zone prediction block for including in fisrt feature figure is specifically as follows head zone prediction block
Coordinate information, because having mapping relations between fisrt feature figure and the image of input, this is because the image size of input
Size with the fisrt feature figure of output is fixed, therefore after there is known the coordinate information in fisrt feature figure, Ke Yigen
The corresponding coordinate information in image is determined according to the coordinate information in the fisrt feature figure, is according to the coordinate information in the image
It can determine the subgraph in image comprising head zone.
After the above process obtains fisrt feature figure, according to the location information of the head zone prediction block in fisrt feature figure,
Determine the subgraph comprising head zone of the image comprising human body to be identified.It is specific to determine that method is, according to fisrt feature figure
With the mapping relations of the image comprising human body to be identified, the subgraph comprising head zone is obtained.
S103: the subgraph is input in the second convolution neural network model that training is completed in advance, described in acquisition
It whether include safety cap in subgraph in the head zone for including.
The identification of safety cap is carried out for the subgraph comprising head zone, recognition result is more accurate reasonable, can be true
Determine whether people is worn on safety cap on head, rather than hold in the hand, or certain region being placed in image.
Subgraph comprising head zone is input in the second convolution neural network model, by the convolution sum of several layers
Pondization operation, and finally pass through full articulamentum, whether final determine in the subgraph comprising head zone includes safety cap.
In embodiments of the present invention, the image comprising human body to be identified is input to the first volume that training is completed in advance first
In product neural network model, the fisrt feature figure of the image is obtained；According to the position of head zone prediction block in fisrt feature figure
The mapping relations of information and fisrt feature figure and the image comprising human body to be identified determine the subgraph comprising head zone；
Detected for the subgraph comprising head zone, can be determined more accurately the people whether safe wearing cap.To include
The subgraph of head zone is input in the second convolution neural network model that training is completed in advance, includes in acquisition subgraph
It whether include safety cap in head zone.In embodiments of the present invention, the subgraph comprising head zone is determined in the picture first
Picture, identifies whether in the subgraph later comprising safety cap, so as to accurately judge people whether safe wearing cap, improve
The accuracy of identification.
Embodiment 2:
In order to the image comprising human body to be identified be determined more accurately the position of head zone, in above-described embodiment
On the basis of, it is in embodiments of the present invention, described that the image comprising human body to be identified is input to the first volume that training is completed in advance
Before in product neural network model, the method also includes:
Described image is zoomed into the first default size.
Since a convolutional neural networks model is when handling the image of different sizes, obtained result also has difference
It is different, therefore in order to which the head zone location information of the image of human body to be identified is determined more accurately, it is examined to the image
Before survey, first by the image scaling comprising human body to be identified to the first fixed preset size.
A kind of possible implementation method will include that the image of human body to be identified is input to the first convolution neural network model
Before, it carries out crop operation or warp to the image to operate, by the image scaling to the first default size, this is first default big
It is small to be preset, when being trained to the first convolution neural network model, sample image can also be zoomed to this first
Preset size, the first default size can be arbitrary value.For example, by being operated comprising the image of human body to be identified by crop
Or warp operation, zoom to the image of 104*104 size.
Embodiment 3:
In order to more accurately judge the classification results of the subgraph comprising head zone, on the basis of the various embodiments described above
On, it is in embodiments of the present invention, described that the subgraph is input to the second convolution neural network model that training is completed in advance
In before, the method also includes:
The subgraph comprising head zone is zoomed into the second default size.
Since a convolutional neural networks model is when handling the image of different sizes, obtained result also has difference
It is different, therefore the classification results in order to more accurately judge the subgraph comprising head zone, which is being carried out to identify it
Before, the subgraph comprising head zone is zoomed into the preset size of fixed second first.
A kind of possible implementation method will include that the subgraph of head zone is input to the second convolution neural network model
Before, crop operation or warp are carried out to the subgraph to operate, which is zoomed into the second default size, this is second pre-
If size be it is preset, when being trained to the second convolution neural network model, also sample image can be zoomed to this
Second preset size, the second default size can be arbitrary value, and the second default size is generally less than the first default size.Example
Such as, the subgraph comprising head zone is operated by crop or warp is operated, zoom to the image of 64*64 size.
For example, Fig. 2 is the structure chart of above-mentioned the first convolution neural network model and the second convolution neural network model,
Middle left (left and right as shown in the figure) branches into the implementation procedure of the first convolution neural network model, which is used for header area
The detection of domain part.Right (left and right as shown in the figure) branches into the implementation procedure of the second convolution neural network model, the right branch net
Network is used for the classification of head zone, for judging whether head zone has safe wearing cap.Wherein process α only includes several volumes
Lamination and pond layer, the parameter of left branch network and the convolution kernel in right branch network share process α, process β are convolutional layer, mistake
Journey γ is full articulamentum.
It is sized specifically, image A is scaled first, such as this first is sized and can will contract for 104*104
Image A after putting is input to left branch network, respectively after process α and process β, obtains fisrt feature figure, the fisrt feature
The size of figure is 13*13*35, wherein includes the location information of head zone prediction block, the location information in the fisrt feature figure
For finding out the subgraph B comprising head zone.After obtaining the subgraph B comprising head zone, subgraph B is zoomed to the
Two default sizes, such as the second default size can be 64*64, and the subgraph B after scaling is input to right branch network, point
Not after process α and process γ, second feature figure is obtained, wherein the size of second feature figure is 1*1*4, the second feature figure
It wears a safety helmet for exporting to belong to, wear common hat, helmet and bareheaded label and confidence level, confidence level highest pair
The label answered is the classification results of the image.
In specific implementation, the specific operation process of process α is as shown in table 1, wherein conv1, conv2, conv3,
Conv4, conv5, conv6 represent convolution operation, and the size of convolution kernel is respectively set as 3*3,3*3,1*1,3*3,3*3,1*
1.Pool1, pool2, pool3 represent pondization operation, and convolution kernel size is set as 2*2.Inner is full articulamentum,
Stride represents distance when doing convolution operation between adjacent convolution kernel, and padding represents when carrying out convolution needs to figure
As the pixel number being filled, A is the image of the first convolution neural network model input, and B is the second convolution neural network model
The image of input.
The size of image A is 104*104, and the characteristic pattern 1 of 16 104*104 is obtained after first layer convolution operation, will
This feature Fig. 1 obtains the characteristic pattern 2 of 16 52*52 after the operation of first layer pondization, and this feature Fig. 2 is passed through second layer convolution
The characteristic pattern 3 of 64 52*52 is obtained after operation, and so on, when finally obtaining 35 13*13 after layer 6 convolution operation
Fisrt feature figure, wherein the purpose of layer 6 convolution is the dimension in order to reduce feature, dimension, that is, port number, by a 1*
1 convolution makes the characteristic pattern of 128 dimensions be reduced to 35 dimensions.
The size of image B is 64*64, after the layer operation of several layers convolution sum pond, when by third layer pond pool3
After operation, the characteristic pattern of 128 8*8 is obtained, the characteristic pattern of 4 1*1 is finally obtained by connecting a full articulamentum.4 1*1
Characteristic pattern be 4 confidence levels, respectively correspond to belong to and wear a safety helmet, helmet, wearing common hat and bareheaded set
Reliability.The corresponding label of confidence level highest is the recognition result of head zone, according to confidence level obtain people whether safe wearing
Cap.
Table 1
Embodiment 4:
In order to carry out head zone detection to the image comprising human body to be identified, therefore also wrapped before being detected to it
Pre-training process is included, on the basis of the various embodiments described above, in embodiments of the present invention, trains first convolutional Neural in advance
The process of network model includes:
Sample image is obtained, wherein being labelled with the location information of head zone frame in the sample image；
Each sample image is input in the first convolution neural network model, according to the first convolution neural network model
Each output is trained the first convolution neural network model.
Since the purpose of the first convolution neural network model is the head zone in order to detect the people in images to be recognized,
Image to be identified is input in the first convolution neural network model be head zone in the image in order to obtain position
Information.Therefore before being trained to the first convolution neural network model, it is necessary first to be labeled to image, obtain sample
This image.Specifically, being labelled with the location information of head zone in each sample image.
Specifically, selecting YOLO as first convolution before being trained to the first convolution neural network model
The loss function of neural network model.Specifically, the calculation method of YOLO loss function is shown below:Wherein, x, y are the centre coordinate value of prediction block, w, h
The respectively width and height of prediction block, c are classification, and p (c) is the probability value of certain corresponding class.
In specific implementation, a certain number of batch sample images are inputted every time, using propagated forward, backpropagation and weight
Update and etc. the parameter of model is updated.Specific training process: the initialization of network progress weight；Input sample figure
As obtaining output valve by the propagation forward of convolutional layer, pond layer；Find out the error between the output valve of network and target value；When
When error must it is expected greater than us, error is passed back in network, the error of each layer is successively found out.The error of each layer can be understood as
For the overall error of network, network should bear how many；When error is equal to or less than desired value, training terminates；It is missed according to acquiring
Difference carries out right value update.It continually enters batch sample image and repeats above step, continuous adjusting parameter, corrective networks output and phase
The error of prestige value finally obtains the network parameter of optimization, i.e. the first convolution neural network model of training completion.
Embodiment 5:
It in order to be identified to the subgraph comprising head zone, therefore further include pre-training before being identified to it
Process in embodiments of the present invention, trains the second convolution neural network model on the basis of the various embodiments described above in advance
Process include:
Sample image is obtained, wherein be labelled with the corresponding label information of the sample image in the sample image, wherein institute
State label information include wear a safety helmet label, wear common hat label, helmet label and label of not puing a label on；
Each sample image is input in the second convolution neural network model, according to the second convolution neural network model
Each output is trained the second convolution neural network model.
Due to the purpose of the second convolution neural network model be in order to comprising head zone subgraph realize classification,
Subgraph comprising head zone is input in the second convolution neural network model be the image in order to obtain classification knot
Fruit.Therefore before being trained to the second convolution neural network model, it is necessary first to indicate to image, obtain sample
Image.Specifically, being labelled with the tag along sort of the image in each sample image, wherein include altogether label in 4, be respectively
Wear a safety helmet label, wear common hat label, helmet label and label of not puing a label on.Specifically, to second convolution mind
Before being trained through network model, select Softmax loss as the loss function of the second convolution neural network model.
Specifically, the calculation method of Softmax loss function is shown below:Wherein, L is loss, SjIt is
J-th of value of the output vector S of Softmax, T are classification number, and y is the vector of a 1*T.
In specific implementation, a certain number of batch sample images are inputted every time, using propagated forward, backpropagation and weight
Update and etc. the parameter of model is updated.Specific training process: the initialization of network progress weight；Input sample figure
As obtaining output valve by the propagation forward of convolutional layer, pond layer；Find out the error between the output valve of network and target value；When
When error must it is expected greater than us, error is passed back in network, the error of each layer is successively found out.The error of each layer can be understood as
For the overall error of network, network should bear how many；When error is equal to or less than desired value, training terminates；It is missed according to acquiring
Difference carries out right value update.It continually enters batch sample image and repeats above step, continuous adjusting parameter, corrective networks output and phase
The error of prestige value finally obtains the network parameter of optimization, i.e. the second convolution neural network model of training completion.
Embodiment 6:
In order to save the time-consuming of identification classification results, the accuracy rate of classification and identification is improved, in the base of the various embodiments described above
On plinth, in embodiments of the present invention, the first convolution neural network model and the second convolution neural network model part are identical.
The embodiment of the present invention is under the premise of guaranteeing recognition effect, in order to save the time-consuming of identification classification results, to packet
On the basis of image containing human body to be identified carries out the detection of head zone, and Classification and Identification is carried out to the subgraph of head zone
Operation.That is the ginseng of the first convolution neural network model and the second convolution neural network model community network structure and shared convolution kernel
Number.
In specific implementation, after the completion of being trained using the above method to the first convolution neural network model, the second convolution
The network structure and convolution nuclear parameter for the first convolution neural network model that neural network model is completed using training.By the first volume
The last layer convolutional layer and loss layer of product neural network model replace with a full articulamentum and Softmax loss function i.e.
It can.Without carrying out the training of the second convolution neural network model, the time-consuming of identification is saved.
It is now described in detail with a kind of detailed process of the specific embodiment to safety cap recognition methods, such as Fig. 3
Shown, which includes that characteristic extraction procedure and safety cap wear identification process.First from the image comprising human body to be identified
The area zoom of human body is extracted to 104*104, the image after scaling is input to the left branch of the network N in Fig. 2, is passed through
α process and β process in Fig. 2, obtain the characteristic pattern of 13*13*35, and this feature figure includes that the position of head zone prediction block is believed
Breath, obtains the coordinate in original image by the mapping relations of characteristic pattern and original image, then takes out from the image of human body and measure in advance
The head zone image arrived.The head zone image scaling is input to 64*64 to the right branch of the network N in Fig. 2, passes through figure
α process and γ process in 2, obtain belonging to and wear a safety helmet, helmet, wearing common hat and bareheaded confidence level, set
The highest corresponding label of reliability is the recognition result of head zone, according to confidence level obtain people whether safe wearing cap.
Embodiment 7:
Fig. 4 is a kind of safety cap identification device provided in an embodiment of the present invention, which includes:
First obtains module 401, for will include that the image of human body to be identified is input to the first volume that training is completed in advance
In product neural network model, the fisrt feature figure of described image is obtained, wherein the fisrt feature figure includes head zone prediction
The location information of frame；
Determining module 402, for according to the location information of head zone prediction block in fisrt feature figure and described first
The mapping relations of characteristic pattern and described image determine the subgraph in described image comprising head zone；
Second obtains module 403, for the subgraph to be input to the second convolutional neural networks that training is completed in advance
In model, whether obtain in the head zone for including in the subgraph comprising safety cap.
Described device further include:
First Zoom module 404, for described image to be zoomed to the first default size.
Described device further include:
Second Zoom module 405, for the subgraph comprising head zone to be zoomed to the second default size.
Described device further include:
First training module 406, for obtaining sample image, wherein being labelled with head zone prediction in the sample image
The location information of frame；Each sample image is input in the first convolution neural network model, according to the first convolutional neural networks
Each output of model is trained the first convolution neural network model.
Described device further include:
Second training module 407, for obtaining sample image, wherein being labelled with the sample image pair in the sample image
The label information answered, wherein the label information include wear a safety helmet label, wear common hat label, helmet and label and do not wear
Helmet label；Each sample image is input in the second convolution neural network model, according to the second convolution neural network model
Each output, the second convolution neural network model is trained.
In conclusion the embodiment of the present invention provides a kind of safety cap recognition methods and device, comprising: will include people to be identified
The image of body is input in the first convolution neural network model that training is completed in advance, obtains the fisrt feature figure of described image,
Wherein, the fisrt feature figure includes the location information of head zone prediction block；It is predicted according to head zone in fisrt feature figure
The mapping relations of the location information of frame and the fisrt feature figure and described image determine in described image comprising header area
The subgraph in domain；The subgraph is input in the second convolution neural network model that training is completed in advance, obtains the son
It whether include safety cap in image in the head zone for including.
Due in embodiments of the present invention, the image comprising human body to be identified being input to the first volume that training is completed in advance
In product neural network model, the fisrt feature figure of the image is obtained；According to the position of head zone prediction block in fisrt feature figure
The mapping relations of information and fisrt feature figure and the image comprising human body to be identified determine the subgraph comprising head zone；
Detected for the subgraph comprising head zone, can be determined more accurately the people whether safe wearing cap.To include
The subgraph of head zone is input in the second convolution neural network model that training is completed in advance, includes in acquisition subgraph
Whether include safety cap in head zone, in embodiments of the present invention, determines the subgraph comprising head zone in the picture first
Picture, identifies whether in the subgraph later comprising safety cap, so as to accurately judge people whether safe wearing cap, improve
The accuracy of identification.
Above-mentioned computer readable storage medium can be any usable medium that the processor in electronic equipment can access
Or data storage device, including but not limited to magnetic storage such as floppy disk, hard disk, tape, magneto-optic disk (MO) etc., optical memory
Such as CD, DVD, BD, HVD and semiconductor memory such as ROM, EPROM, EEPROM, nonvolatile memory (NAND
FLASH), solid state hard disk (SSD) etc..
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Although preferred embodiments of the present invention have been described, it is created once a person skilled in the art knows basic
Property concept, then additional changes and modifications may be made to these embodiments.So it includes excellent that the following claims are intended to be interpreted as
It selects embodiment and falls into all change and modification of the scope of the invention.
Obviously, various changes and modifications can be made to the invention without departing from essence of the invention by those skilled in the art
Mind and range.In this way, if these modifications and changes of the present invention belongs to the range of the claims in the present invention and its equivalent technologies
Within, then the present invention is also intended to include these modifications and variations.
Claims (11)
1. a kind of safety cap recognition methods, which is characterized in that the described method includes:
Image comprising human body to be identified is input in the first convolution neural network model that training is completed in advance, described in acquisition
The fisrt feature figure of image, wherein the fisrt feature figure includes the location information of head zone prediction block；
According to reflecting for the location information of head zone prediction block in fisrt feature figure and the fisrt feature figure and described image
Relationship is penetrated, determines the subgraph in described image comprising head zone；
The subgraph is input in the second convolution neural network model that training is completed in advance, obtains and wrapped in the subgraph
It whether include safety cap in the head zone contained.
2. the method as described in claim 1, which is characterized in that described that the image comprising human body to be identified is input to preparatory instruction
Before practicing in the first convolution neural network model completed, the method also includes:
Described image is zoomed into the first default size.
3. the method as described in claim 1, which is characterized in that that the subgraph is input to training in advance is completed
Before in two convolutional neural networks models, the method also includes:
The subgraph comprising head zone is zoomed into the second default size.
4. the method as described in claim 1, which is characterized in that train the process of the first convolution neural network model in advance
Include:
Sample image is obtained, wherein being labelled with the location information of head zone frame in the sample image；
Each sample image is input in the first convolution neural network model, according to each of first convolution neural network model
Output, is trained the first convolution neural network model.
5. the method as described in claim 1, which is characterized in that train the process of the second convolution neural network model in advance
Include:
Sample image is obtained, wherein the corresponding label information of the sample image is labelled in the sample image, wherein the mark
Label information include wear a safety helmet label, wear common hat label, helmet label and label of not puing a label on；
Each sample image is input in the second convolution neural network model, according to each of second convolution neural network model
Output, is trained the second convolution neural network model.
6. the method according to claim 1 to 5, which is characterized in that the first convolution neural network model and second
Convolutional neural networks model part is identical.
7. a kind of safety cap identification device, which is characterized in that described device includes:
First obtains module, for will include that the image of human body to be identified is input to the first convolution nerve net that training is completed in advance
In network model, the fisrt feature figure of described image is obtained, wherein the fisrt feature figure includes the position of head zone prediction block
Information；
Determining module, for according to the location information of head zone prediction block in fisrt feature figure and the fisrt feature figure
With the mapping relations of described image, the subgraph in described image comprising head zone is determined；
Second obtains module, for the subgraph to be input in the second convolution neural network model that training is completed in advance,
It whether obtains in the head zone for including in the subgraph comprising safety cap.
8. device as claimed in claim 7, which is characterized in that described device further include:
First Zoom module, for described image to be zoomed to the first default size.
9. device as claimed in claim 7, which is characterized in that described device further include:
Second Zoom module, for the subgraph comprising head zone to be zoomed to the second default size.
10. device as claimed in claim 7, which is characterized in that described device further include:
First training module, for obtaining sample image, wherein being labelled with the position letter of head zone frame in the sample image
Breath；Each sample image is input in the first convolution neural network model, according to each of first convolution neural network model
Output, is trained the first convolution neural network model.
11. device as claimed in claim 7, which is characterized in that described device further include:
Second training module, for obtaining sample image, wherein being labelled with the corresponding mark of the sample image in the sample image
Sign information, wherein the label information include wear a safety helmet label, wear common hat label, helmet label and mark of not puing a label on
Label；Each sample image is input in the second convolution neural network model, according to each of second convolution neural network model
Output, is trained the second convolution neural network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811570198.1A CN109753898A (en) | 2018-12-21 | 2018-12-21 | A kind of safety cap recognition methods and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811570198.1A CN109753898A (en) | 2018-12-21 | 2018-12-21 | A kind of safety cap recognition methods and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109753898A true CN109753898A (en) | 2019-05-14 |
Family
ID=66403894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811570198.1A Pending CN109753898A (en) | 2018-12-21 | 2018-12-21 | A kind of safety cap recognition methods and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109753898A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110222672A (en) * | 2019-06-19 | 2019-09-10 | 广东工业大学 | The safety cap of construction site wears detection method, device, equipment and storage medium |
| CN110458075A (en) * | 2019-08-05 | 2019-11-15 | 北京泰豪信息科技有限公司 | Detection method, storage medium, detection device and the detection system that safety cap is worn |
| CN110619314A (en) * | 2019-09-24 | 2019-12-27 | 杭州宇泛智能科技有限公司 | Safety helmet detection method and device and electronic equipment |
| CN111401278A (en) * | 2020-03-20 | 2020-07-10 | 重庆紫光华山智安科技有限公司 | Helmet identification method and device, electronic equipment and storage medium |
| CN111814762A (en) * | 2020-08-24 | 2020-10-23 | 深延科技(北京)有限公司 | Helmet wearing detection method and device |
| US11048948B2 (en) * | 2019-06-10 | 2021-06-29 | City University Of Hong Kong | System and method for counting objects |
| CN113191273A (en) * | 2021-04-30 | 2021-07-30 | 西安聚全网络科技有限公司 | Oil field well site video target detection and identification method and system based on neural network |
| CN113256934A (en) * | 2021-05-18 | 2021-08-13 | 哈尔滨理工大学 | Safety detection system for aerial work personnel |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106372662A (en) * | 2016-08-30 | 2017-02-01 | 腾讯科技（深圳）有限公司 | Helmet wearing detection method and device, camera, and server |
| CN107103617A (en) * | 2017-03-27 | 2017-08-29 | 国机智能科技有限公司 | The recognition methods of safety cap wearing state and system based on optical flow method |
| CN107527053A (en) * | 2017-08-31 | 2017-12-29 | 北京小米移动软件有限公司 | Object detection method and device |
| CN107679524A (en) * | 2017-10-31 | 2018-02-09 | 天津天地伟业信息系统集成有限公司 | A kind of detection method of the safety cap wear condition based on video |
| CN108647619A (en) * | 2018-05-02 | 2018-10-12 | 安徽大学 | The detection method and device that safety cap is worn in a kind of video based on deep learning |
-
2018
- 2018-12-21 CN CN201811570198.1A patent/CN109753898A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106372662A (en) * | 2016-08-30 | 2017-02-01 | 腾讯科技（深圳）有限公司 | Helmet wearing detection method and device, camera, and server |
| CN107103617A (en) * | 2017-03-27 | 2017-08-29 | 国机智能科技有限公司 | The recognition methods of safety cap wearing state and system based on optical flow method |
| CN107527053A (en) * | 2017-08-31 | 2017-12-29 | 北京小米移动软件有限公司 | Object detection method and device |
| CN107679524A (en) * | 2017-10-31 | 2018-02-09 | 天津天地伟业信息系统集成有限公司 | A kind of detection method of the safety cap wear condition based on video |
| CN108647619A (en) * | 2018-05-02 | 2018-10-12 | 安徽大学 | The detection method and device that safety cap is worn in a kind of video based on deep learning |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11048948B2 (en) * | 2019-06-10 | 2021-06-29 | City University Of Hong Kong | System and method for counting objects |
| CN110222672A (en) * | 2019-06-19 | 2019-09-10 | 广东工业大学 | The safety cap of construction site wears detection method, device, equipment and storage medium |
| CN110458075A (en) * | 2019-08-05 | 2019-11-15 | 北京泰豪信息科技有限公司 | Detection method, storage medium, detection device and the detection system that safety cap is worn |
| CN110458075B (en) * | 2019-08-05 | 2023-08-25 | 北京泰豪信息科技有限公司 | Method, storage medium, device and system for detecting wearing of safety helmet |
| CN110619314A (en) * | 2019-09-24 | 2019-12-27 | 杭州宇泛智能科技有限公司 | Safety helmet detection method and device and electronic equipment |
| CN111401278A (en) * | 2020-03-20 | 2020-07-10 | 重庆紫光华山智安科技有限公司 | Helmet identification method and device, electronic equipment and storage medium |
| CN111814762A (en) * | 2020-08-24 | 2020-10-23 | 深延科技(北京)有限公司 | Helmet wearing detection method and device |
| CN113191273A (en) * | 2021-04-30 | 2021-07-30 | 西安聚全网络科技有限公司 | Oil field well site video target detection and identification method and system based on neural network |
| CN113256934A (en) * | 2021-05-18 | 2021-08-13 | 哈尔滨理工大学 | Safety detection system for aerial work personnel |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109753898A (en) | A kind of safety cap recognition methods and device | |
| Shi et al. | Landslide recognition by deep convolutional neural network and change detection | |
| Aboah et al. | Real-time multi-class helmet violation detection using few-shot data sampling technique and yolov8 | |
| CN106127204B (en) | A kind of multi-direction meter reading Region detection algorithms of full convolutional neural networks | |
| CN108830188A (en) | Vehicle checking method based on deep learning | |
| CN111723786B (en) | Method and device for detecting wearing of safety helmet based on single model prediction | |
| CN110245662A (en) | Detection model training method, device, computer equipment and storage medium | |
| CN110147743A (en) | Real-time online pedestrian analysis and number system and method under a kind of complex scene | |
| CN109117876A (en) | A kind of dense small target deteection model building method, model and detection method | |
| CN113269073A (en) | Ship multi-target tracking method based on YOLO V5 algorithm | |
| CN109657592A (en) | A kind of face identification system and method for intelligent excavator | |
| CN109241871A (en) | A kind of public domain stream of people's tracking based on video data | |
| CN109543596A (en) | A kind of water level monitoring method, apparatus, electronic equipment and storage medium | |
| CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
| CN110414559A (en) | The construction method and commodity recognition method of intelligence retail cabinet commodity target detection Unified frame | |
| CN106909886A (en) | A kind of high accuracy method for traffic sign detection and system based on deep learning | |
| CN105160355B (en) | A kind of method for detecting change of remote sensing image based on region correlation and vision word | |
| CN107977656A (en) | A kind of pedestrian recognition methods and system again | |
| CN110728252B (en) | Face detection method applied to regional personnel motion trail monitoring | |
| CN109858547A (en) | A kind of object detection method and device based on BSSD | |
| CN110321885A (en) | A kind of acquisition methods and device of point of interest | |
| CN109522963A (en) | A kind of the feature building object detection method and system of single-unit operation | |
| CN108805016A (en) | A kind of head and shoulder method for detecting area and device | |
| CN110322472A (en) | A kind of multi-object tracking method and terminal device | |
| CN109389105A (en) | A kind of iris detection and viewpoint classification method based on multitask |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190514 |
|
| RJ01 | Rejection of invention patent application after publication |