CN109299641A - A kind of train dispatcher's fatigue monitoring image adaptive Processing Algorithm - Google Patents
A kind of train dispatcher's fatigue monitoring image adaptive Processing Algorithm Download PDFInfo
- Publication number
- CN109299641A CN109299641A CN201810354996.4A CN201810354996A CN109299641A CN 109299641 A CN109299641 A CN 109299641A CN 201810354996 A CN201810354996 A CN 201810354996A CN 109299641 A CN109299641 A CN 109299641A
- Authority
- CN
- China
- Prior art keywords
- face
- image
- region
- eye
- human eye
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/16—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state
- A61B5/18—Devices for psychotechnics; Testing reaction times ; Devices for evaluating the psychological state for vehicle drivers or machine operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/18—Eye characteristics, e.g. of the iris
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/18—Eye characteristics, e.g. of the iris
- G06V40/197—Matching; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Ophthalmology & Optometry (AREA)
- Psychiatry (AREA)
- Pathology (AREA)
- Educational Technology (AREA)
- Hospice & Palliative Care (AREA)
- Child & Adolescent Psychology (AREA)
- Psychology (AREA)
- Social Psychology (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Biophysics (AREA)
- Developmental Disabilities (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of train dispatcher's fatigue monitoring image adaptive Processing Algorithms, belong to the technical field of the pattern-recognition based on biological characteristic, by adaptive Face datection algorithm by previous frame image detection result optimizing current frame image detection parameters, detection range is reduced to greatest extent, reduction process detects number, improves image detection efficiency；It according to face human eye relationship and eyes positional relationship, is detected by adaptive fast human-eye and intelligent estimation algorithm, further reduces human eye detection range, while carried out eye position and effectively inferring and data check, effectively improve data accuracy and completeness；Consecutive image testing result is assessed into subsequent a period of time picture quality according to interval recognition of face and frame-skipping quick Processing Algorithm, make differentiation frame-skipping processing, improve image processing efficiency, subsequent processes are adaptively adjusted according to currently processed acquisition data by self-adapting detecting technology with reaching, can be improved the purpose of quality of image processing and efficiency.
Description
Technical field
The invention belongs to the technical fields of the pattern-recognition based on biological characteristic, are related to image procossing, pattern-recognition, calculating
Many-sided theoretical and technology such as machine vision and Human physiology, in particular to a kind of train dispatcher's fatigue monitoring image
Self-adaptive processing algorithm.
Background technique
Face datection, recognition of face and human eye detection are to determine face and human eye in the picture according to face and eye feature
Region, the corresponding object identity of identification face, it is more to be related to image procossing, pattern-recognition, computer vision and Human physiology etc.
Aspect theory and technology.
OpenCV (Open Source Computer Vision Library) is that Intel company initiates and participates in developing
Cross-platform computer vision library, be made of a series of C functions and a small amount of C++ class, realize image procossing and computer view
Feel many general-purpose algorithms of aspect.OpenCV can be operated in Linux, Windows and Mac OS operating system, be provided simultaneously
The language interfaces such as Python, Ruby and MATLAB, have cross-platform, lightweight and efficiently, independent of other external libraries and
Free open source feature, is the ideal tools of image procossing, pattern-recognition and computer vision field secondary development.
OpenCV provides the basic library dll of numerous image procossings, pattern-recognition and computer vision field basic function, but
Be its significant drawback be almost without provide GUI interface, it is difficult to directly meet the needs of application development.
EmguCV is then the cross-platform .Net encapsulation of one of OpenCV, allows directly to be adjusted with .Net language by encapsulation
With OpenCV function, C# and OpenCV can be connected well, to make up deficiency of the OpenCV in terms of GUI.
Face datection and human eye detection belong to object detection field, and cascade AdaBoost algorithm is that OpenCV is supported and wide
The algorithm of target detection of general application obtains AdaBoost cascade classifier using the training of sample Haar feature, calls Haar detection
Function realizes target detection.Wherein AdaBoost algorithm core concept is the weak typing different for the training of the same training set
These Weak Classifier collection adaptives are promoted to strong classifier by device, are weighted final convergence by iteration and are tended towards stability.
Image detection and image recognition technology are quickly grown, and Face datection and face recognition technology are extensive in all trades and professions
Using.Contactless PERCLOS method based on human eye closure degree obtains industry and is widely recognized as, and starts in driver
It is applied in flyer's fatigue monitoring, achieves good result.At the same time, EMGUCV provides the system of convenient and efficient
Interface can be realized Face datection and recognition of face basic function, meet basic need.
Train dispatcher's dispatch control working environment has significant open, compass of competency is wide, integrated information is wide, equipment and
System is more, and technical equipment arrangement generallys use multiple rows of multiple row mode.Mainly sight is focused primarily on other industry personnel
To difference, train dispatcher can focus different zones in different periods according to need of work during dispatch control in front,
It is likely to occur new line, bows or the movement such as left and right side view, focus vision have significant dispersing characteristic.
According to dispatch control operative scenario feature, human body physiological characteristics and fatigue monitoring needs, train dispatcher's fatigue prison
Examining system performance requirement is mainly reflected in Noninvasive, concurrency, continuity, high efficiency and accuracy totally five aspects.Image
Processing is fatigue monitoring link the most time-consuming, is related to fatigue monitoring system data-handling efficiency and performance, train dispatcher
Parallel fatigue monitoring proposes requirements at the higher level to image processing techniques efficiency, and the prior art has been difficult to meet needs.
Summary of the invention
In order to solve the above problems existing in the present technology, it is an object of that present invention to provide a kind of train dispatcher's fatigue prisons
Altimetric image self-adaptive processing algorithm meets system in quality and speed side to reach the function realization in image procossing disparate modules
The performance requirement in face adaptively adjusts subsequent processes according to currently processed acquisition data by self-adapting detecting technology
It is whole, quality of image processing and efficiency can be improved to the maximum extent, meet system performance while realizing image processing function
The purpose of demand.
The technical scheme adopted by the invention is as follows: a kind of train dispatcher's fatigue monitoring image adaptive Processing Algorithm, base
In the development platform of VS2010, EmguCV is called to carry out secondary development using C# language, mainly include the following:
(1) Face datection and human eye detection
FaceHaar is obtained by load face classification device haarcascade_frontalface_alt2.xml, by adding
Manned eye classifier haarcascade_mcs_righteye.xml obtains EyeHaar, calls DetectMultiScale function
Respectively obtain following formula:
Faces=FaceHaar.DetectMultiScale (Image1, SF1, MinNH1, MinSize1, MaxSize1) (2-1)
Eyes=EyeHaar.DetectMultiScale (Image2, SF2, MinNH2, MinSize2, MaxSize2) (2-2)
Wherein, DetectMultiScale is the multi-dimension testing method of CascadeClassifier class, obtains input figure
The regional ensemble of specific objective object as in；
Image1 and Image2 respectively indicates the image object of Face datection and human eye detection, and type is Image < Gray,
byt e>；
SF1 and SF2 respectively indicates the zoom factor of Face datection and human eye detection；
MinNH1 and MinNH2 respectively indicates the minimum number for constituting the adjacent rectangle of Face datection and human eye detection target；
MinSize1 and MaxSize1 respectively indicates the minimum dimension and full-size that Face datection obtains rectangular area；
MinSize2 and MaxSize2 respectively indicates the minimum dimension and full-size that human eye detection obtains rectangular area；
(2) recognition of face
Recognition of face is by calling the Recognize method of EigenObjectRecognizer class in EmguCV to realize, base
In the human face region that Face datection obtains, by the identity of face characteristic discrimination objective object, face recognition process traversal is current
The human face region that frame image Face datection obtains, until finding the human face region for belonging to target object, then carries out subsequent people
Eye detection and eyelid distance computation, the formula of critical process are as follows:
Recognizer=newEigenObjectRecognizer (Images, Labels, DistanceThreshold,
termCrit) (2-3)
Name=recognizer.Recognize (result) .Label (2-4)
Images be face recognition training pattern matrix, type be Image<Gray, byte>；
Labels is the corresponding identification number array of recognition of face pattern matrix, type string；
DistanceThreshold is characterized distance threshold；
TermCrit is face recognition training standard, type MCvTermCriteria；
Name is the object identity mark that recognition of face obtains, and belongs to element in Labels.
Further, the Face datection uses the quick self-adapted Face datection algorithm constrained based on interframe, in face
In detection zone, Face datection search window carries out sequence detection since MinSize1 size, if it cannot detect face
Search window expands SF1 times, is recycled and is carried out until detecting face or until search window size reaches MaxSize1 with this；It enables
I is the frame variable of image procossing, PRiFor the image rectangular area, DRiFor the image Face datection target area, FRiFor the figure
As the face rectangular area detected, then:
MinSize1i≤FRi.Size≤MaxSize1i (2-6)
The Face datection target area for taking next frame image is DRi+1, window size MinSize1i+1With
MaxSize1i+1, enable f1、f2And f3Respectively indicate DRi+1、MinSize1i+1And MaxSize1i+1With FRiBetween auto-adaptive function
Relationship:
DRi+1=f1(FRi)1≤i≤M-1,i∈N (2-7)
MinSize1i+1=f2(FRi)1≤i≤M-1,i∈N (2-8)
MaxSize1i+1=f3(FRi)1≤i≤M-1,i∈N (2-9)
Wherein, M is the number of image frames of current video file.
Further, enabling λ is that coefficient is expanded in region of search, then Face datection target area DR in i+1 frame imagei+1's
Location parameter is X and Y, and dimensional parameters are Width and Height；The f1Auto-adaptive function is indicated using following formula:
α and β is enabled to respectively indicate MinSize1i+1And MaxSize1i+1Relative to FRiThe scaling of size, then function f2With
f3Formula (2-11) and (2-12), which can be respectively adopted, to be indicated:
Further, when being likely to occur DR during atual detectioni+1Beyond PRi+1The case where, wherein PRi+1It is next
The image rectangular area of frame image need to be carried out according to the actual situation by Face datection target area DRi+1It is modified to feasible
DRi′+1；Take DRi+1With PRi+1Intersection is as i+1 frame image Face datection target area, then DRi′+1=DRi+1∩PRi+1。
Further, the human eye detection uses adaptive fast human-eye detection algorithm, enables as ERiThe i-th frame of video file
FR is based in imageiThe human eye detection target area determined with face " three five, front yards " rules self-adaptive, human eye detection target area
Domain ERiLocation parameter be X and Y, dimensional parameters be Width and Height, determine ERiWith FRiBetween auto-adaptive function relationship
It is as follows:
Further according to human eye detection region ERiAdaptively determine human eye detection minimum search window MinSize2iMost wantonly search for
Rope window MaxSize2i, and MinSize2iAnd MaxSize2iWith ERiBetween the following formula of auto-adaptive function relationship:
Further, eye position deduction and data check are carried out by self-adapting intelligent algorithm under specific circumstances, had
Body is as follows: enabling LERiAnd RERiRespectively in ERiDetected left-eye image region and eye image region, q frame in atmosphere
It includes: LER that image human eye self-adapting intelligent, which is inferred and verified referring to information,p、RERp、FRpAnd FRq, wherein p be q frame image with
Before detect complete human eye information and face information frame number variable maximum value, p≤q-1；
Enable ERNpFor ERpDirect detected human eye quantity in range, q frame image human eye is intelligently in deduction and verification
Hold because of ERNpIt is worth difference and difference, specifically includes following three kinds of scenes:
(1) if ERNp>=2, according to LERp、RERp、FRpAnd FRqInformation verifies detected eye areas one by one, rejects
Retain two best eye areas after extra eye areas, relationship determines left-eye image region LER respectively depending on the relative positionqWith
Eye image region RERq；
(2) if ERNp=1, according to LERp、RERp、FRpAnd FRqInformation carries out eye areas verification, determines that directly detection obtains
The eye areas obtained is left eye region LERqOr right eye region RERq, by examine after based on this eye areas in ERqModel
Enclose interior deduction another eye areas；
(3) if ERNp=0, according to LERp、RERp、FRpAnd FRqInformation is directly in ERqLeft eye region LER is inferred in rangeq
With right eye region RERq。
Further, according to LERp、RERp、FRpAnd FRqThe eyes region LER that self-adapting intelligent is inferredq' and RERq' respectively
Pass through following formula:
Wherein, sq,pZoom factor for human eye area in q frame image relative to human eye area in pth frame image,
sq,p=(FRq.Width/FRp.Width+FRq.Height/FRp.Height)/2(2-18)。
Further, mesh is checked according to face location parameter and dimensional parameters situation of change by interval face recognition algorithms
The identity for marking object calls the Recognize method of EigenObjectRecognizer class in EmguCV true after condition triggering
Determine the corresponding personnel identity of human face region, the touching for needing to carry out recognition of face after facial image is detected in i+1 frame image
Clockwork spring part includes:
(1) the previous frame image of current frame image fails to detect human face region, i.e. DRi+1=PRi, indicate to detect
Human face region is the new facial area into personnel in image range；
(2) the face rectangular area FR detected in i+1 frame imagei+1It cannot meet simultaneously following formula:
Wherein, 0 < ω≤0.4,0 < σ≤0.15, FR is takeniFor the face rectangular area detected in the i-th frame image,
FRi+1For the face rectangular area detected in i+1 frame image.
Further, when target object, which leaves video record, takes the photograph range, Face datection can be carried out during image processing
Frame-skipping processing, the trigger condition of Face datection frame-skipping processing are set as continuous K frame image and fail to detect human face region, and parameter
The value range of K is [5,25].
The invention has the benefit that
1. Face datection and human eye detection based on EmguCV have good robustness, in posture slight shift and target
It remains to accurately detect face and human eye area when partial occlusion；
2. by using quick self-adapted Face datection algorithm, according to the face location and ruler detected in sequential frame image
Very little data are adaptively adjusted the detection zone and relevant parameter of Face datection, maximum on the basis of ensuring detection accuracy
Improve detection rates to limit；
3. in the human eye region of search determined based on human face region, adaptive fast human-eye detection and intelligence infer algorithm energy
It is enough that corresponding eyes region is found to robustness under different situations, eyes eyelid distance is obtained after binocular images regional processing, is led to
Eyes region completion in image is crossed to improve the accuracy and robustness of human eye detection；
4. solving the problems, such as that the object identity after Face datection is checked using interval face recognition technology differentiation, ensuring
Image processing efficiency can be improved while image process target accuracy to the maximum extent；
5. target object leaves Face datection when range is taken the photograph in video record and carries out frame-skipping by using frame-skipping quick Processing Algorithm
Processing, image overall treatment efficiency can be effectively improved without Face datection when range is taken the photograph in video record by leaving in target object.
Detailed description of the invention
Fig. 1 be image rectangular area in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention,
Face datection target area and face rectangular area relation schematic diagram；
Fig. 2 be in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention target object along X-axis
The mobile face rectangular area in direction changes schematic diagram；
Fig. 3 be in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention target object along Z axis
The mobile face rectangular area in direction changes schematic diagram；
Fig. 4 be in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention target object along X-axis
Change schematic diagram with the mobile face rectangular area of Z-direction；
Fig. 5 is target object Y-axis side in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention
Face regional change schematic diagram is moved forwards, backwards；
Fig. 6 is same under different distance in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention
Differentiation embodies schematic diagram in the picture for size and displacement；
Fig. 7 is people in different modes in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention
The comprehensive time-consuming lateral comparison schematic diagram of face detection；
Fig. 8 is facial image " three in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention
Five, front yard " distribution schematic diagram；
Fig. 9 is in train dispatcher's fatigue monitoring image adaptive Processing Algorithm provided by the invention at Face datection frame-skipping
Triggering, continuous trigger and the normal process schematic of recovery of reason.
1--FRiFor the face rectangular area that the image detection arrives, 2--DRiFor the image Face datection target area, 3--
PRiFor the image rectangular area, O-- video capture device, 4--FF mode, 5--AF mode, 6--FA mode, 7--AA mode,
8--BS mode, 9-- ear, 10-- eyes, 11-- nose, 12-- mouth.
Specific embodiment
With reference to the accompanying drawing and specific embodiment the present invention is further elaborated.
The present invention provides a kind of train dispatcher's fatigue monitoring image adaptive Processing Algorithm, the exploitation based on VS2010
Platform calls EmguCV to carry out secondary development, mainly using C# language include the following:
(1) Face datection and human eye detection
FaceHaar is obtained by load face classification device haarcascade_frontalface_alt2.xml,
FaceHaar is the Face datection example of AdaBoost cascade classifier CascadeClassifier, passes through load human eye classification
Device haarcascade_mcs_righteye.xml obtains EyeHaar, and EyeHaar is AdaBoost cascade classifier
The human eye detection example of CascadeClassifier calls DetectMultiScale function to respectively obtain following formula:
Faces=FaceHaar.DetectMultiScale (Image1, SF1, MinNH1, MinSize1, MaxSize1)
(2-1)
Eyes=EyeHaar.DetectMultiScale (Image2, SF2, MinNH2, MinSize2, MaxSize2)
(2-2)
Wherein, Faces is the human face region rectangular array that image Face datection returns, and type is Rectangle [], yuan
Element includes the positions and dimensions information of a face；
Eyes is the human eye area rectangular array that image human eye detection returns, and type is Rectangle [], and element includes
The positions and dimensions information of one eye eyeball；
DetectMultiScale is the multi-dimension testing method of CascadeClassifier class, is obtained in input picture
The regional ensemble of specific objective object；
Image1 and Image2 respectively indicates the image object of Face datection and human eye detection, and type is Image < Gray,
Byt e >, according to the geometry inclusion relation of eyes and facial area, human eye detection in the human face region that Face datection obtains into
Row, i.e. Image2 is the corresponding image-region of Faces element, if face cannot be detected, without subsequent human eye detection；
SF1 and SF2 respectively indicates the zoom factor of Face datection and human eye detection, indicates to search in the adjacent scanning twice of front and back
Rope window proportionality coefficient, default value are that the 1.1 each search windows of expression successively expand 10%, specifically can voluntarily be set as needed
It is fixed；
MinNH1 and MinNH2 respectively indicates the minimum number for constituting the adjacent rectangle of Face datection and human eye detection target
Min_neighbors, the small rectangle number of composition detection target and less than min_neighbors when, can all be excluded, default value
It is 3.If value is 0, representative function does not do any operation and is returned to all tested candidate rectangles, and the setting is commonly used in use
The customized combinator to testing result in family；
MinSize1 and MaxSize1 respectively indicate Face datection obtain rectangular area minimum dimension and full-size, two
For person's synergy to limit the range of human face region, type is Size；
MinSize2 and MaxSize2 respectively indicate human eye detection obtain rectangular area minimum dimension and full-size, two
For person's synergy to limit the range of human eye area, type is Size；
(2) recognition of face
Recognition of face is by calling the Recognize method of EigenObjectRecognizer class in EmguCV to realize, base
In the human face region that Face datection obtains, by the identity of face characteristic discrimination objective object, face recognition process traversal is current
The human face region that frame image Face datection obtains, until finding the human face region for belonging to target object, then carries out subsequent people
Eye detection and eyelid distance computation, the formula of critical process are as follows:
Recognizer=newEigenObjectRecognizer (Images, Labels, DistanceThreshold,
termCrit) (2-3)
Name=recognizer.Recognize (result) .Label (2-4)
EigenObjectRecognizer uses the target marker of PCA；
Recognizer is an example of EigenObjectRecognizer class, and Recognize method can obtain
Special object identification information；
Images be face recognition training pattern matrix, type be Image<Gray, byte>, each image size just as
And normalized by histogram, Images is obtained by artificial training in advance；
Labels is the corresponding identification number array of recognition of face pattern matrix, and type string, element and training are schemed
As for image in the presence of the mapping relations being corresponding in turn to, Labels is corresponding specified in Images training in array；
DistanceThreshold is characterized distance threshold, and the value is bigger, then recognition of face is more difficult to but accuracy of identification is got over
It is high；
TermCrit is face recognition training standard, type MCvTermCriteria；
Name is the object identity mark that recognition of face obtains, and belongs to element in Labels.
When target object workplace has open nature, the possible of short duration video that leaves is adopted in the target object course of work
Collect region, other target objects may also appear in video collection area.Therefore for acquire video a certain frame image and
Speech, Face datection are likely to occur three kinds of results: (1) without human face region；(2) human face regions, belong to target target object or
Other target objects of person；(3) multiple human face regions including target target object or do not include that target target object exists
It is interior.
Recognition of face passes through face characteristic discrimination objective object identity based on the human face region that Face datection obtains.
Face recognition process traverses the human face region that current frame image Face datection obtains, and the people of target target object is belonged to until finding
Until face region, then carry out subsequent human eye detection and eyelid distance computation.
The eyelid spacing data that recognition of face can ensure that image procossing obtains is to belong to specific target object, Yi Mianying
Ring the data accuracy of its degree of fatigue development.
Face datection basic function realize after, need further satisfaction fatigue monitoring rate and in terms of performance need
It asks, the Face datection is used to the quick self-adapted Face datection algorithm constrained based on interframe, i.e., according to aforementioned successive frame figure
The face location and dimension data detected as in, is adaptively adjusted the detection zone and relevant parameter of Face datection, true
Detection rates are improved to the maximum extent on the basis of guarantor's detection accuracy.
In Face datection region, Face datection search window carries out sequence detection since MinSize1 size, if not
It can detect that then search window expands SF1 times to face, progress is recycled until detecting that face or search window size reach with this
Until MaxSize1；Enabling i is the frame variable of image procossing, PRiFor the image rectangular area, DRiFor the image Face datection target
Region, FRiFor the face rectangular area that the image detection arrives, PRiImage rectangular area, DRiFace datection target area and FRi
Relationship between face rectangular area is as shown in Figure 1, then:
MinSize1i≤FRi.Size≤MaxSize1i (2-6)
It is 25 frames/s that face video, which acquires frame frequency, and consecutive frame image time interval is 0.04s, face location parameter and size
Parameters variation all has gradually changeable, and this change procedure is subtly portrayed by sequential frame image record.Single-frame images people
Face testing result can directly reflect that face location and dimension information, the Face datection result of sequential frame image then further contain
The variation tendency of face location and size, provide effective reference for next frame image Face datection.
With the detected face location parameter of sequential frame image and dimensional parameters information, next frame figure is adaptively determined
As Face datection region DRi+1Location parameter and dimensional parameters, MinSize1i+1And MaxSize1i+1, by accurately determining people
Face detection zone position minimizes DRi+1Size and MaxSize1i+1, maximize MinSize1i+1, face inspection is reduced to greatest extent
It surveys region and reduces detection number, and then promote Face datection rate.
The Face datection target area for taking next frame image is DRi+1, window size MinSize1i+1With
MaxSize1i+1, when face is not detected in the i-th frame image, DRi+1、MinSize1i+1And MaxSize1i+1Take initial value silent
Recognize value；When the i-th frame image detection is to face, then i+1 frame image Face datection parameters are according to FRiAdaptively really
It is fixed, enable f1、f2And f3Respectively indicate DRi+1、MinSize1i+1And MaxSize1i+1With FRiBetween auto-adaptive function relationship:
DRi+1=f1(FRi)1≤i≤M-1,i∈N (2-7)
MinSize1i+1=f2(FRi)1≤i≤M-1,i∈N (2-8)
MaxSize1i+1=f3(FRi)1≤i≤M-1,i∈N (2-9)
Wherein, M is the number of image frames of current video file.
Enabling λ is that coefficient is expanded in region of search, then Face datection target area DR in i+1 frame imagei+1Location parameter be
X and Y, dimensional parameters are Width and Height；The f1Auto-adaptive function is indicated using following formula:
When being likely to occur DR during atual detectioni+1Beyond PRi+1The case where, wherein PRi+1For next frame image
Image rectangular area need to be carried out according to the actual situation by Face datection target area DRi+1It is modified to feasible DRi′+1, i.e.,Take DRi+1With PRi+1Intersection is as i+1 frame image Face datection target area, then DRi′+1=DRi+1∩
PRi+1。
Preferably, taking λ=0.4 in above-mentioned formula, make a concrete analysis of as follows:
Target object position can occur because of need of work to from left to right (X-direction), forward backward (Y direction) or upwards to
Under (Z-direction) displacement, be displaced some direction for being likely to occur in three directions, it is also possible to two of them direction or
Three directions.The position of video capture device is to maintain changeless during video acquisition, and target object displacement will lead to
It acquires the face location obtained or corresponding change occurs for size.X-axis will lead to face location generation with Z-direction variation and mutually strain
Change, corresponding human face region possibly is present in all ranges of the positive and negative maximum displacement of both direction, respectively such as Fig. 2, Fig. 3 and Fig. 4
It is shown.
Target object can influence human face region positions and dimensions in image, target object in Y direction back-and-forth motion simultaneously
It moves forward, face is closer apart from camera, and corresponding facial image area size is bigger；Conversely, face is remoter apart from camera,
Corresponding facial image size is smaller.Target object Y direction is moved forward and backward schematic diagram as shown in figure 5, target object facial dimension
Camera distance change is N times, then the length of the facial image rectangular area detected and wide variation are 1/N times.
Movement speed is about 1m/s for each person under normal circumstances, and change in location process record is in continuous 25 frame image in 1s
In, the actual average displacement maximum value of adjacent two field pictures septum reset is 4cm or so.For the same target object, face
Portion's actual size be to maintain it is constant, apart from video camera distance it is remoter, the facial area that Face datection obtains is smaller, same journey
The variation of the actual displacement reflection of degree in the picture is smaller；Conversely, the facial area that Face datection obtains is bigger, equal extent
The variation of actual displacement reflection in the picture is bigger.As shown in fig. 6, OE=2OA, same size area exists at ABCD and EFGH
A is presented as in image respectively1B1C1D1And E1F1G1H1, same scale is displaced is presented as A respectively in the picture1A2And E1E2, wherein
A1B1=2E1F1, A1A2=2E1E2。
Target object face actual size and displacement are objective reality, and human face region size and displacement are in the picture
Size, which synchronizes, to be zoomed in or out, and the face location and size detected using current frame image determines next frame Face datection
Region has significant adaptive and higher efficiency.By target object work top height and widths affect, target object distance
Video capture device distance will not be less than 40cm, therefore Y direction is moved forward and backward the people for causing to detect in adjacent two field pictures
Face area size change rate is not more than 10%.Face average-size is about 11cm*18cm, then face exists in adjacent two field pictures
X-direction and Z-direction displacement are usually less than the 40% of face width.From the point of view of comprehensive three direction of displacement, present frame face area
Face width 40% is expanded in four direction in domain outward respectively can be used as next frame image Face datection region under normal conditions.
It is found based on train dispatcher's Y direction position and the analytical calculation for being moved forward and backward speed, it is same in consecutive frame image
The variation of one facial size does not exceed 10%, therefore FR substantiallyiDR is determined adaptive when can geti+1While can also be
FRi+1Size provides reference, passes through minimum search window MinSize1i+1With maximum search window MaxSize1i+1People is described respectively
FR in face detection processi+1The lower and upper limit of size, utilize FRiAdaptively maximize MinSize1i+1And minimum
MaxSize1i+1FR can be reduced to the maximum extenti+1Size feasible region can effectively improve detection speed.
α and β is enabled to respectively indicate MinSize1i+1And MaxSize1i+1Relative to FRiThe scaling of size, then function f2With
f3Formula (2-11) and (2-12), which can be respectively adopted, to be indicated:
On the basis of comprehensively considering consecutive frame facial size amplitude of variation, α and β more while leveling off to 1, i+1 frame figure
Picture Face datection speed is faster, considers 5% surplus capacity on the basis of 10% scaling, it is preferred that value is 0.85 and 1.15 respectively.
Carry out the experiment of different mode human face detection time: random selection Sample video carries out Face datection test, will
650 frame video images are divided into 13 groups, and every group of 50 frame sequentials carry out image-capture, pretreatment and Face datection, under different mode
The Face datection time is as shown in Figure 7.
Wherein, AA indicates DRi+1、MinSize1i+1And MaxSize1i+1It is based on FRiIt is adaptive to determine；AF is indicated
MinSize1i+1And MaxSize1i+1Based on FRiIt is adaptive to determine, DRi+1For universe range (DRi+1=PRi+1)；FF is indicated
MinSize1i+1And MaxSize1i+1For fixed value, DRi+1For universe range；FA indicates MinSize1i+1And MaxSize1i+1For
Fixed value, DRi+1Based on FRiIt is adaptive to determine；BS indicates based process mode, only carries out frame image-capture and pretreatment.
The average time difference that AA, AF, FF, FA and B/S mode carry out Face datection is huge, 50 frame image procossing mean times
Between be respectively 317ms, 435ms, 724ms, 405ms and 217ms.It is FF pattern systhesis speed that AA mode human face, which detects overall rate,
Twice or more of rate removes the based process such as image-capture, image preprocessing and analytical calculation work (B/S mode content), individually
Face datection part rate under AA mode is 5 times or more of rate under FF mode, it can be seen that, adaptive Face datection efficiency
Promote significant effect.
Human eye detection is the basic premise of subsequent eye closure degree judgement in image processing process, usually to detect
Human face region is human eye detection range, improves human eye detection rate by reducing detection range with this.Face face organ is empty
Between be distributed the special ratios relationship for usually meeting " three five, front yards ", wherein three front yards refer to the length ratio of face, the length point of face
It is forehead hairline line respectively to brow ridge, brow ridge to nose bottom and nose bottom to lower chin for three equal parts；Five refer to the width ratio of face
Example, is divided into five eye-shaped length for the face width from left side hairline to right side hairline, eyes lateral position is located at second
A and the 4th eye-shaped extension position, as shown in Figure 8, wherein ear 9 is up to eyebrow down toward pen tip；Eyes 10 in face 1/2
Place；11 bottom of nose is at the 1/2 of the centre of eyes and chin, and width is the interval width of two eyes, and mouth 12 is in nose
At 11 and the 1/3 of chin.
Based on face " three five, front yards " space constraint relationship, human eye inspection further can be adaptively reduced within the scope of face
Survey range, under different location, different posture and different scale human eye detection zone as adaptive change occurs for human face region,
And detected eyes region is completely included, the human eye detection uses adaptive fast human-eye detection algorithm, enables as ERiDepending on
FR is based in frequency file the i-th frame imageiThe human eye detection target area determined with face " three five, front yards " rules self-adaptive, human eye
Detect target area ERiLocation parameter be X and Y, dimensional parameters be Width and Height, determine ERiWith FRiBetween it is adaptive
Answer functional relation as follows:
Based on size relationship between human eye and face, human eye detection search window is carried out using Sample video and is tested
Card, further according to human eye detection region ERiAdaptively determine human eye detection minimum search window MinSize2iWith maximum search window
Mouth MaxSize2i, detection rates, and MinSize2 can be farthest improved while guaranteeing accuracy in detectioniWith
MaxSize2iWith ERiBetween the following formula of auto-adaptive function relationship:
Eye position deduction and data check are carried out under specific circumstances by self-adapting intelligent algorithm, and human eye detection is adaptive
It should intelligently infer and checking algorithm is using human eye and face physiological characteristic as theoretical basis, specifically include:
(1) stability of the position and scale relativeness of human eye and face；
(2) the almost the same property of eyes size；
(3) synchronism that face and human eye change on position and scale；
(4) eyes closure degree and the synchronism of time；
Above-mentioned human eye and face physiological characteristic rule contain in the video image of acquisition, pass through detected face information
It is emerged from human eye information, therefore directly detected face and human eye information (position and scale) are subsequent human eye detections
Self-adapting intelligent is inferred and the direct basis of verification.
It is specific as follows: to enable LERiAnd RERiRespectively in ERiDetected left-eye image region and eye image in atmosphere
Region, it includes: LER that q frame image human eye self-adapting intelligent, which is inferred and verified referring to information,p、RERp、FRpAnd FRq, wherein p be
The maximum value of the frame number variable of complete human eye information and face information, p≤q-1 are detected before q frame image；
Enable ERNpFor ERpDirect detected human eye quantity in range, q frame image human eye is intelligently in deduction and verification
Hold because of ERNpIt is worth difference and difference, specifically includes following three kinds of scenes:
(1) if ERNp>=2, according to LERp、RERp、FRpAnd FRqInformation verifies detected eye areas one by one, rejects
Retain two best eye areas after extra eye areas, relationship determines left-eye image region LER respectively depending on the relative positionqWith
Eye image region RERq；
(2) if ERNp=1, according to LERp、RERp、FRpAnd FRqInformation carries out eye areas verification, determines that directly detection obtains
The eye areas obtained is left eye region LERqOr right eye region RERq, by examine after based on this eye areas in ERqModel
Enclose interior deduction another eye areas；
(3) if ERNp=0, according to LERp、RERp、FRpAnd FRqInformation is directly in ERqLeft eye region LER is inferred in rangeq
With right eye region RERq。
Self-adapting intelligent is inferred under different scenes and verification content has differences, but it is intelligently inferred and verification principle exists
Substantially it is identical, is to calculate the LER in eyes region using the effective face of previous frame and binocular information as referenceq′
.X、LERq′.Y、LERq′.Width、LERq' .Height and RERq′.X、RERq′.Y、RERq′.Width、RERq' .Height,
Eyes region is estimated in the human face region of current frame image with above-mentioned supplemental characteristic, with this to the human eye area directly detected
It is verified and is inferred.
It is specific as follows:
According to LERp、RERp、FRpAnd FRqThe eyes region LER that self-adapting intelligent is inferredq' and RERq' respectively by as follows
Formula:
Wherein, sq,pZoom factor for human eye area in q frame image relative to human eye area in pth frame image,
sq,p=(FRq.Width/FRp.Width+FRq.Height/FRp.Height)/2 (2-18)。
According to above-mentioned, in ERqAfter detecting eyes in range, the eye areas and LER that detect by comparingq' and
RERq' between position and scaling relation verified, location estimating is carried out to the eye areas that does not directly detect, passes through figure
The completion of eyes region improves the accuracy and robustness of human eye detection as in.
Target object is checked according to face location parameter and dimensional parameters situation of change by interval face recognition algorithms
Identity, using train dispatcher as target object, the opening in Train Dispatch & Command place causes to be likely to occur in frame image more
The face of a train dispatcher or non-targeted train dispatcher, face recognition technology can check personnel's body by face characteristic
Part, train scheduling platform dispatcher on duty is determined from the human face region detected, the interference of other train dispatchers is eliminated with this.
Recognition of face is carried out for the human face region that each frame image Face datection obtains, object identity can be accurately determined the most,
But image processing work load can be dramatically increased simultaneously, reduces the arrangement processing speed of fatigue monitoring system.
The human face region position of consecutive frame image and size have roll-off characteristic, and have the corresponding changing ratio upper limit.
Carrying out after recognition of face determines object identity, for it is subsequent can detect the sequential frame image of face for, can be according to people
Face position and change in size situation determine object identity.Therefore to determine the recognition of face of object identity in image processing process
In there is no the necessity carried out frame by frame, identified i.e. when according to face location and change in size situation object identity cannot be checked
It can.
Interval face recognition algorithms are setting recognition of face trigger conditions, are called in EmguCV after condition triggering
The Recognize method of EigenObjectR ecognizer class determines the corresponding personnel identity of human face region, in i+1 frame figure
It is detected as in and needs after facial image the trigger condition for carrying out recognition of face to include:
(1) the previous frame image of current frame image fails to detect human face region, i.e. DRi+1=PRi, indicate to detect
Human face region is the new facial area into personnel in image range；
(2) the face rectangular area FR detected in i+1 frame imagei+1It cannot meet simultaneously following formula:
Wherein, 0 < ω≤0.4,0 < σ≤0.15, FR is takeniFor the face rectangular area detected in the i-th frame image,
FRi+1For the face rectangular area detected in i+1 frame image, formula (2-19) is with adaptive under aforementioned interframe constraint condition
Based on answering Face datection to analyze, human face region is public by the condition that the identity based on face location and change in size situation checks
Formula.Wherein, in value range, ω and σ value is smaller, is got over according to the condition that face location and change in size situation check identity
Strictly, the frame amount of images for needing to carry out recognition of face is more.
In scheduler routine commander's course of work, only has dispatch control people on duty before most time train scheduling platforms
Member solves the problems, such as that the object identity after Face datection is checked to interval face recognition technology differentiation, is ensuring image procossing pair
Image processing efficiency can be improved while as accuracy to the maximum extent.
In addition, train dispatcher of short duration may leave video record during dispatch control takes the photograph range, lead to continuous one
Frame image in the section time cannot detect face.According to the adaptive Face datection algorithm under interframe constraint, present frame cannot
Next frame Face datection range is extended to image whole region when detecting face, and the single frames Face datection time is caused significantly to increase
Add.Train dispatcher leave video record the case where taking the photograph range would generally certain time, for this section of time acquisition image
Face datection belongs to invalidation, while when single-frame images processing time appears in frame image range compared with train dispatcher grows
Very much.
Train dispatcher's video record take the photograph frame per second be 25 frames/second, train dispatcher leave video record take the photograph range when without
Face datection can effectively improve image overall treatment efficiency, when target object, which leaves video record, takes the photograph range, in image procossing
Face datection frame-skipping processing can be carried out in the process, and the trigger condition of Face datection frame-skipping processing is set as continuous K frame image and fails
Detect human face region, and the value range of parameter K is [5,25].
Face datection trigger frame-skipping processing when, continuously skip frame number can according to image procossing it needs to be determined that, can [100,
250] 25 fixed integer times is taken in range, corresponding reality time span is 4~10s, is obtained to train dispatcher's degree of fatigue
It will not have an impact.When Face datection frame-skipping processing is triggered by continuous several times, it can sequentially be gradually increased and continuously skip frame number,
But it is not easy more than 1000 frames, the triggering of Face datection frame-skipping, continuous trigger and to restore normal processes as shown in Figure 9.
The present invention is not limited to above-mentioned optional embodiment, anyone can show that other are various under the inspiration of the present invention
The product of form, however, make any variation in its shape or structure, it is all to fall into the claims in the present invention confining spectrum
Technical solution, be within the scope of the present invention.
Claims (9)
1. a kind of train dispatcher's fatigue monitoring image adaptive Processing Algorithm, which is characterized in that the exploitation based on VS2010 is flat
Platform calls EmguCV to carry out secondary development, mainly using C# language include the following:
(1) Face datection and human eye detection
FaceHaar is obtained by load face classification device haarcascade_frontalface_alt2.xml, by loading people
Eye classifier haarcascade_mcs_righteye.xml obtains EyeHaar, calls DetectMultiScale function difference
Obtain following formula:
Faces=FaceHaar.DetectMultiScale (Image1, SF1, MinNH1, MinSize1, MaxSize1) (2-1)
Eyes=EyeHaar.DetectMultiScale (Image2, SF2, MinNH2, MinSize2, MaxSize2) (2-2)
Wherein, DetectMultiScale is the multi-dimension testing method of CascadeClassifier class, is obtained in input picture
The regional ensemble of specific objective object；
Image1 and Image2 respectively indicates the image object of Face datection and human eye detection, and type is Image < Gray, byte
>；
SF1 and SF2 respectively indicates the zoom factor of Face datection and human eye detection；
MinNH1 and MinNH2 respectively indicates the minimum number for constituting the adjacent rectangle of Face datection and human eye detection target；
MinSize1 and MaxSize1 respectively indicates the minimum dimension and full-size that Face datection obtains rectangular area；
MinSize2 and MaxSize2 respectively indicates the minimum dimension and full-size that human eye detection obtains rectangular area；
(2) recognition of face
Recognition of face is based on people by calling the Recognize method of EigenObjectRecognizer class in EmguCV to realize
The detected human face region of face, by the identity of face characteristic discrimination objective object, face recognition process traverses present frame figure
As the human face region that Face datection obtains, until finding the human face region for belonging to target object, then subsequent human eye inspection is carried out
It surveys and eyelid distance computation, the formula of critical process is as follows:
Recognizer=newEigenObjectRecognizer (Images, Labels, DistanceThreshold,
termCrit) (2-3)
Name=recognizer.Recognize (result) .Label (2-4)
Images be face recognition training pattern matrix, type be Image<Gray, byte>；
Labels is the corresponding identification number array of recognition of face pattern matrix, type string；
DistanceThreshold is characterized distance threshold；
TermCrit is face recognition training standard, type MCvTermCriteria；
Name is the object identity mark that recognition of face obtains, and belongs to element in Labels.
2. train dispatcher's fatigue monitoring image adaptive Processing Algorithm according to claim 1, which is characterized in that described
Face datection uses the quick self-adapted Face datection algorithm constrained based on interframe, and in Face datection region, Face datection is searched
Rope window carries out sequence detection since MinSize1 size, and search window expands SF1 times if it cannot detect face, with this
Circulation carries out until detecting that face or search window size reach MaxSize1；Enabling i is the frame variable of image procossing,
PRiFor the image rectangular area, DRiFor the image Face datection target area, FRiThe face rectangle region arrived for the image detection
Domain, then:
MinSize1i≤FRi.Size≤MaxSize1i (2-6)
The Face datection target area for taking next frame image is DRi+1, window size MinSize1i+1And MaxSize1i+1, enable
f1、f2And f3Respectively indicate DRi+1、MinSize1i+1And MaxSize1i+1With FRiBetween auto-adaptive function relationship:
DRi+1=f1(FRi) 1≤i≤M-1,i∈N (2-7)
MinSize1i+1=f2(FRi) 1≤i≤M-1,i∈N (2-8)
MaxSize1i+1=f3(FRi) 1≤i≤M-1,i∈N (2-9)
Wherein, M is the number of image frames of current video file.
3. train dispatcher's fatigue monitoring image adaptive Processing Algorithm according to claim 2, which is characterized in that enable λ
Coefficient is expanded for region of search, then Face datection target area DR in i+1 frame imagei+1Location parameter be X and Y, size ginseng
Number is Width and Height；The f1Auto-adaptive function is indicated using following formula:
α and β is enabled to respectively indicate MinSize1i+1And MaxSize1i+1Relative to FRiThe scaling of size, then function f2And f3It can
Formula (2-11) and (2-12), which is respectively adopted, to be indicated:
4. train dispatcher's fatigue monitoring image adaptive Processing Algorithm according to claim 3, which is characterized in that when
DR is likely to occur during actually detectedi+1Beyond PRi+1The case where, wherein PRi+1For the image rectangular area of next frame image,
It need to carry out Face datection target area DR according to the actual situationi+1It is modified to feasible DR 'i+1；Take DRi+1With PRi+1Intersection is made
For i+1 frame image Face datection target area, then DR 'i+1=DRi+1∩PRi+1。
5. train dispatcher's fatigue monitoring image adaptive Processing Algorithm according to claim 1, it is characterised in that ground, institute
Human eye detection is stated using adaptive fast human-eye detection algorithm, is enabled as ERiFR is based in video file the i-th frame imageiAnd face
The human eye detection target area that " three five, front yards " rules self-adaptive determines, human eye detection target area ERiLocation parameter be X and
Y, dimensional parameters are Width and Height, determine ERiWith FRiBetween auto-adaptive function relationship it is as follows:
Further according to human eye detection region ERiAdaptively determine human eye detection minimum search window MinSize2iWith maximum search window
Mouth MaxSize2i, and MinSize2iAnd MaxSize2iWith ERiBetween the following formula of auto-adaptive function relationship:
6. train dispatcher's fatigue monitoring image adaptive Processing Algorithm according to claim 5, which is characterized in that pass through
Self-adapting intelligent algorithm carries out eye position deduction and data check under specific circumstances, specific as follows: to enable LERiAnd RERiRespectively
For in ERiDetected left-eye image region and eye image region in atmosphere, q frame image human eye self-adapting intelligent are inferred
It include: LER with verifying referring to informationp、RERp、FRpAnd FRq, wherein p be detect before q frame image complete human eye information and
The maximum value of the frame number variable of face information, p≤q-1；
Enable ERNpFor ERpDirect detected human eye quantity in range, q frame image human eye intelligently infer and verification content because
ERNpIt is worth difference and difference, specifically includes following three kinds of scenes:
(1) if ERNp>=2, according to LERp、RERp、FRpAnd FRqInformation verifies detected eye areas one by one, and it is extra to reject
Retain two best eye areas after eye areas, relationship determines left-eye image region LER respectively depending on the relative positionqAnd right eye
Image-region RERq；
(2) if ERNp=1, according to LERp、RERp、FRpAnd FRqInformation carries out eye areas verification, determines directly detected
Eye areas is left eye region LERqOr right eye region RERq, by examine after based on this eye areas in ERqIn range
Infer another eye areas；
(3) if ERNp=0, according to LERp、RERp、FRpAnd FRqInformation is directly in ERqLeft eye region LER is inferred in rangeqAnd right eye
Region RERq。
7. train dispatcher's fatigue monitoring image adaptive Processing Algorithm according to claim 6, which is characterized in that according to
LERp、RERp、FRpAnd FRqThe eyes region LER ' that self-adapting intelligent is inferredqWith RER 'qPass through following formula respectively:
Wherein, sq,pZoom factor for human eye area in q frame image relative to human eye area in pth frame image,
sq,p=(FRq.Width/FRp.Width+FRq.Height/FRp.Height)/2 (2-18)。
8. train dispatcher's fatigue monitoring image adaptive Processing Algorithm according to claim 1, which is characterized in that pass through
It is spaced the identity that face recognition algorithms check target object according to face location parameter and dimensional parameters situation of change, is touched in condition
The Recognize method of EigenObjectRecognizer class in EMGUCV is called to determine the corresponding personnel of human face region after hair
Identity detects in i+1 frame image and needs after facial image the trigger condition for carrying out recognition of face to include:
(1) the previous frame image of current frame image fails to detect human face region, i.e. DRi+1=PRi, indicate the face area detected
Domain is the new facial area into personnel in image range；
(2) the face rectangular area FR detected in i+1 frame imagei+1It cannot meet simultaneously following formula:
Wherein, 0 < ω≤0.4,0 < σ≤0.15, FR is takeniFor the face rectangular area detected in the i-th frame image, FRi+1It is
The face rectangular area detected in i+1 frame image.
9. train dispatcher's fatigue monitoring image adaptive Processing Algorithm according to claim 1, which is characterized in that work as mesh
Mark object leaves video record when taking the photograph range, carries out Face datection frame-skipping processing during image processing, at Face datection frame-skipping
The trigger condition of reason is set as continuous K frame image and fails to detect human face region, and the value range of parameter K is [5,25].
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810354996.4A CN109299641B (en) | 2018-04-19 | 2018-04-19 | Train dispatcher fatigue monitoring image adaptive processing algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810354996.4A CN109299641B (en) | 2018-04-19 | 2018-04-19 | Train dispatcher fatigue monitoring image adaptive processing algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109299641A true CN109299641A (en) | 2019-02-01 |
| CN109299641B CN109299641B (en) | 2020-10-16 |
Family
ID=65167538
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810354996.4A Active CN109299641B (en) | 2018-04-19 | 2018-04-19 | Train dispatcher fatigue monitoring image adaptive processing algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109299641B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111294524A (en) * | 2020-02-24 | 2020-06-16 | 中移（杭州）信息技术有限公司 | Video editing method and device, electronic equipment and storage medium |
| CN112733570A (en) * | 2019-10-14 | 2021-04-30 | 北京眼神智能科技有限公司 | Glasses detection method and device, electronic equipment and storage medium |
| CN113505674A (en) * | 2021-06-30 | 2021-10-15 | 上海商汤临港智能科技有限公司 | Face image processing method and device, electronic equipment and storage medium |
| CN114821747A (en) * | 2022-05-26 | 2022-07-29 | 深圳市科荣软件股份有限公司 | Method and device for identifying abnormal state of construction site personnel |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101593425A (en) * | 2009-05-06 | 2009-12-02 | 深圳市汉华安道科技有限责任公司 | A kind of fatigue driving monitoring method and system based on machine vision |
| CN101599207A (en) * | 2009-05-06 | 2009-12-09 | 深圳市汉华安道科技有限责任公司 | A kind of fatigue driving detection device and automobile |
| CN104408878A (en) * | 2014-11-05 | 2015-03-11 | 唐郁文 | Vehicle fleet fatigue driving early warning monitoring system and method |
| CN104866843A (en) * | 2015-06-05 | 2015-08-26 | 中国人民解放军国防科学技术大学 | Monitoring-video-oriented masked face detection method |
| CN107491769A (en) * | 2017-09-11 | 2017-12-19 | 中国地质大学（武汉） | Method for detecting fatigue driving and system based on AdaBoost algorithms |
-
2018
- 2018-04-19 CN CN201810354996.4A patent/CN109299641B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101593425A (en) * | 2009-05-06 | 2009-12-02 | 深圳市汉华安道科技有限责任公司 | A kind of fatigue driving monitoring method and system based on machine vision |
| CN101599207A (en) * | 2009-05-06 | 2009-12-09 | 深圳市汉华安道科技有限责任公司 | A kind of fatigue driving detection device and automobile |
| CN104408878A (en) * | 2014-11-05 | 2015-03-11 | 唐郁文 | Vehicle fleet fatigue driving early warning monitoring system and method |
| CN104866843A (en) * | 2015-06-05 | 2015-08-26 | 中国人民解放军国防科学技术大学 | Monitoring-video-oriented masked face detection method |
| CN107491769A (en) * | 2017-09-11 | 2017-12-19 | 中国地质大学（武汉） | Method for detecting fatigue driving and system based on AdaBoost algorithms |
Non-Patent Citations (2)
| Title |
|---|
| IT屋: "EmguCV - 人脸识别 - 使用Microsoft Access数据库的训练集", 《IT屋》 * |
| 匿名者2: "OpenCV人脸识别--detectMultiScale函数", 《博客园》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112733570A (en) * | 2019-10-14 | 2021-04-30 | 北京眼神智能科技有限公司 | Glasses detection method and device, electronic equipment and storage medium |
| CN112733570B (en) * | 2019-10-14 | 2024-04-30 | 北京眼神智能科技有限公司 | Glasses detection method and device, electronic equipment and storage medium |
| CN111294524A (en) * | 2020-02-24 | 2020-06-16 | 中移（杭州）信息技术有限公司 | Video editing method and device, electronic equipment and storage medium |
| CN111294524B (en) * | 2020-02-24 | 2022-10-04 | 中移（杭州）信息技术有限公司 | Video editing method and device, electronic equipment and storage medium |
| CN113505674A (en) * | 2021-06-30 | 2021-10-15 | 上海商汤临港智能科技有限公司 | Face image processing method and device, electronic equipment and storage medium |
| CN114821747A (en) * | 2022-05-26 | 2022-07-29 | 深圳市科荣软件股份有限公司 | Method and device for identifying abnormal state of construction site personnel |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109299641B (en) | 2020-10-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299641A (en) | A kind of train dispatcher's fatigue monitoring image adaptive Processing Algorithm | |
| EP3767522A1 (en) | Image recognition method and apparatus, and terminal and storage medium | |
| US6879709B2 (en) | System and method for automatically detecting neutral expressionless faces in digital images | |
| WO2018188453A1 (en) | Method for determining human face area, storage medium, and computer device | |
| CN105868716B (en) | A kind of face identification method based on facial geometric feature | |
| CN109635727A (en) | A kind of facial expression recognizing method and device | |
| CN110443189A (en) | Face character recognition methods based on multitask multi-tag study convolutional neural networks | |
| CN109657583A (en) | Face's critical point detection method, apparatus, computer equipment and storage medium | |
| CN109325462B (en) | Face recognition living body detection method and device based on iris | |
| CN112784763A (en) | Expression recognition method and system based on local and overall feature adaptive fusion | |
| Linder et al. | Real-time full-body human gender recognition in (RGB)-D data | |
| CN109858375A (en) | Living body faces detection method, terminal and computer readable storage medium | |
| CN109389002A (en) | Biopsy method and device | |
| CN110084192A (en) | Quick dynamic hand gesture recognition system and method based on target detection | |
| CN109086659A (en) | A kind of Human bodys' response method and apparatus based on multimode road Fusion Features | |
| CN109977867A (en) | A kind of infrared biopsy method based on machine learning multiple features fusion | |
| CN109063626A (en) | Dynamic human face recognition methods and device | |
| CN109711309A (en) | A kind of method whether automatic identification portrait picture closes one's eyes | |
| CN111291773A (en) | Feature identification method and device | |
| CN105912126A (en) | Method for adaptively adjusting gain, mapped to interface, of gesture movement | |
| CN108174141A (en) | A kind of method of video communication and a kind of mobile device | |
| Bouhabba et al. | Support vector machine for face emotion detection on real time basis | |
| Yuan et al. | Ear detection based on CenterNet | |
| Yaseen et al. | A Novel Approach Based on Multi-Level Bottleneck Attention Modules Using Self-Guided Dropblock for Person Re-Identification | |
| Ray et al. | Design and implementation of affective e-learning strategy based on facial emotion recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |