CN108932465A - Reduce the method, apparatus and electronic equipment of Face datection false detection rate - Google Patents
Reduce the method, apparatus and electronic equipment of Face datection false detection rate Download PDFInfo
- Publication number
- CN108932465A CN108932465A CN201711462162.7A CN201711462162A CN108932465A CN 108932465 A CN108932465 A CN 108932465A CN 201711462162 A CN201711462162 A CN 201711462162A CN 108932465 A CN108932465 A CN 108932465A
- Authority
- CN
- China
- Prior art keywords
- block
- frame difference
- frame image
- object block
- pixel
- 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
- G06V40/172—Classification, e.g. identification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The embodiment of the invention provides a kind of method, apparatus and electronic equipment for reducing Face datection false detection rate, this method includes：Obtain the previous frame image of current frame image and the current frame image to be detected；Multiple first object blocks are obtained from current frame image, and obtain the second target block in previous frame image with each first object block respective coordinates；Judge whether each first object block is static object according to the frame difference relationship of the second target block of each first object block and corresponding coordinate；The first object block for being determined as static object is deleted from current frame image.By above step, the Moving Objects in the target captured can be accurately judged to using frame difference relationship, environmental diversity interference caused by detection is reduced, improves the accuracy of judgement.
Description
Technical field
The present invention relates to technical field of computer vision, in particular to a kind of side for reducing Face datection false detection rate
Method, device and electronic equipment.
Background technique
The human face detection tech role indispensable in intelligent information epoch performer.Firstly, Face datection is from moving
A key link in face identifying system.Secondly, Face datection is in information retrieval based on contents, Digital Image Processing, video detection
And safety monitoring etc. has important application value.With the development of intellectual technology, method for detecting human face is also continuous
Innovation, from based on critical point detection and matched conventional method to the deep learning method based on CNN, Face datection accuracy rate
And timeliness is also continuously available promotion.However, either method of the conventional method still based on depth network model, face are examined
The problem of surveying erroneous detection is constantly present.This problem seems abnormal bad for the application as face snap machine etc, because
Capturing system report candid photograph to face while also can constantly report by the target of erroneous detection, especially report fixed before camera lens
A certain fixed erroneous detection object under background.
Currently, solve the problems, such as face erroneous detection have become based on most important problem in depth network model series methods it
One, face detection system is used especially under complex environment background.Due to face detection system usage scenario and illumination etc.
The diversity of environmental factor, it is existing by increasing negative sample to overcome erroneous detection be apparently not completely safe plan, therefore, how to realize
A kind of pair of system is captured accuracy rate and performance and will not be impacted, and the problem of can reducing Face datection accuracy rate urgently solves
Certainly.
Summary of the invention
In view of this, the object of the present invention is to provide a kind of method, apparatus and electronics for reducing Face datection false detection rate
Equipment is to solve the above problems.
Presently preferred embodiments of the present invention provides a kind of method for reducing Face datection false detection rate, the method includes：
Obtain the previous frame image of current frame image to be detected and the current frame image；
Multiple first object blocks are obtained from the current frame image, and obtain in the previous frame image with it is each described
Second target block of first object block respective coordinates；
Each institute is judged according to the frame difference relationship of the second target block of each first object block and corresponding coordinate
State whether first object block is static object；
If it is determined that then the first object block for being determined as static object is deleted from the current frame image for static dynamic
It removes.
Further, the step that the first object block that will be determined as static object is deleted from the current frame image
After rapid, the method also includes：
Remaining first object block in current frame image after progress delete processing is input to the face classification of foundation
In device, to filter out face object from the remaining first object block.
Further, the frame of second target block according to each first object block and corresponding coordinate is poor
Relationship judges the step of whether each first object block is static object, including：
Calculus of differences is carried out to the second target block of each first object block and corresponding coordinate, it is each to obtain
The frame difference figure of the first object block and the second corresponding target block；
For each frame difference figure, the pixel value of each pixel in the frame difference figure is compared with the first preset threshold
Compared with it is equivalent to obtain each pixel corresponding frame difference according to comparison result；
The corresponding first object block of the frame difference figure and the second target are obtained according to the frame difference equivalence of each pixel
Frame difference total value between block；
Judge whether the first object block is static object according to the frame difference total value.
Further, the pixel value by each pixel in the frame difference figure is compared with the first preset threshold, root
The corresponding frame difference equivalence of each pixel is obtained according to comparison result, the frame difference figure is obtained according to the frame difference equivalence of each pixel
The step of frame difference total value between corresponding first object block and the second target block, including：
The frame difference figure is divided into multiple sub-blocks；
For each sub-block, by the pixel value of pixel each in the sub-block respectively with first preset threshold
It is compared；
The frame difference equivalence that pixel value is greater than the pixel of first preset threshold is set to 1, pixel value is less than or equal to
The frame difference equivalence of the pixel of first preset threshold is set to 0；
Count the equivalent summation of the frame difference for the pixel for including in the sub-block；
The summation equivalent to each sub-block corresponding frame difference adds up, to obtain the first object block and institute
State the frame difference total value between the second target block.
Further, it is described according to the frame difference total value judge the first object block whether be static object step
Suddenly, including：
The frame difference total value is detected whether less than the second preset threshold, if being less than second preset threshold, determines institute
Stating first object block is static object；
If it is greater than or equal to second preset threshold, then the frame difference that predetermined number is chosen from the multiple sub-block is equivalent
The maximum sub-block of summation；
The summation equivalent to the frame difference of the sub-block of selection adds up, the accumulated value being calculated and the frame difference total value
Between ratio；
It detects whether the ratio is greater than third predetermined threshold value, if more than the third predetermined threshold value, then determines described
One target block is static object.
Further, the face classification device is obtained by following steps：
Construct the classifier network architecture based on convolutional neural networks；
The multiple facial image positive samples that will acquire and multiple negative samples are separately input into the classifier network architecture
To be trained to the facial image positive sample and the negative sample, to obtain the face classification device.
Further, described to carry out remaining first object block in the current frame image after delete processing and be input to and build
In vertical face classification device, the step of to filter out face object from the remaining first object block, including：
First object block remaining after deletion is input to the face classification device of foundation；
The between facial image positive sample after detecting in the first object block and the face classification device training
Second between negative sample in one degree of fitting and the first object block and the face classification device after training fits
Degree；
Corresponding first degree of fitting of the first object block is compared with the second degree of fitting, if first fitting
Degree is greater than second degree of fitting, then determines the first object block for face object.
Further, described that multiple first object blocks are obtained from the current frame image, and obtain the previous frame
Before the step of the second target block in image with each first object block respective coordinates, the method also includes：
The current frame image and the previous frame image are scaled in proportion.
Another preferred embodiment of the invention also provides a kind of device for reducing Face datection false detection rate, described device packet
It includes：
Image collection module, for obtaining the previous frame figure of current frame image to be detected and the current frame image
Picture；
Target block obtains module, for obtaining multiple first object blocks from the current frame image, and obtains institute
State the second target block in previous frame image with each first object block respective coordinates；
Judgment module, it is poor for the frame according to each first object block and the second target block of corresponding coordinate
Relationship judges whether each first object block is static object；
Removing module will be determined as the first object block of static object from described for when being determined as static object
It is deleted in current frame image.
Another preferred embodiment of the invention also provides a kind of electronic equipment, including：
Memory；
Processor；And
The device of Face datection false detection rate is reduced, including one or more is stored in the memory and by the processing
The software function module that device executes.
The embodiment of the present invention provides a kind of method, apparatus and electronic equipment for reducing Face datection false detection rate, by from working as
Multiple first object blocks are obtained in prior image frame, and are obtained and each first object from the previous frame image of the current frame image
Block has multiple second target blocks of respective coordinates, according to each first object block and the second corresponding target block
Frame difference relationship to judge whether each first object block is static object, and when being determined as static object by corresponding first
Target block is deleted from current frame image.By above step, the target captured can be accurately judged to using frame difference relationship
In Moving Objects, reduce environmental diversity interference caused by detection, improve the accuracy of judgement.It is of the invention to make
Above objects, features, and advantages can be clearer and more comprehensible, and preferred embodiment is cited below particularly, and cooperate appended attached drawing, be elaborated
It is as follows.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, below will be to needed in the embodiment attached
Figure is briefly described, it should be understood that the following drawings illustrates only certain embodiments of the present invention, therefore is not construed as pair
The restriction of range for those of ordinary skill in the art without creative efforts, can also be according to this
A little attached drawings obtain other relevant attached drawings.
Fig. 1 is the structural block diagram for the electronic equipment that present pre-ferred embodiments provide.
Fig. 2 is the flow chart of the method for the reduction Face datection false detection rate that present pre-ferred embodiments provide.
Fig. 3 is the flow chart of the sub-step of step S105 in Fig. 2.
Fig. 4 is the flow chart of the sub-step of step S1053 in Fig. 3.
Fig. 5 is the flow chart of the sub-step of step S1055 in Fig. 3.
Fig. 6 is another flow chart of the method for the reduction Face datection false detection rate that present pre-ferred embodiments provide.
Fig. 7 is the flow chart for the method for establishing face classification device that present pre-ferred embodiments provide.
Fig. 8 is the classifier network architecture schematic diagram constructed in present pre-ferred embodiments.
Fig. 9 is the flow chart of the sub-step of step S109 in Fig. 6.
Figure 10 is the functional block diagram of the device for the reduction Face datection false detection rate that present pre-ferred embodiments provide.
Icon：100- electronic equipment；The device of 110- reduction Face datection false detection rate；111- image collection module；112-
Target block obtains module；113- judgment module；114- removing module；115- screening module；120- processor；130- storage
Device.
Specific embodiment
Inventor it has been investigated that, realized in the prior art frequently with following methods to after face snap target sieve
Choosing：
(1) a kind of mode is in face detection system after face snap link, using the method based on color filter
All targets captured are screened, think non-face target with filtering appts.
The method carries out prospect judgement to all targets captured first, then utilizes trained colour of skin foreground classification
For device pair it is determined that the candid photograph target for prospect carries out marking screening, meet set threshold condition is judged as human face target
And export, the application such as recognition of face for follow-up link does not meet then delete target.
(2) another way is part face detection system after face snap link, using based on template matching and
Similar method carries out matching screening to all targets captured, and thinks non-face target with filtering appts.
Such method carries out facial contour or Partial key point location to all targets captured first, then utilizes thing
A kind of or a few class face template first made and the target carry out the matching judgment of key point, meet set threshold condition
Be judged as face and export target, the application such as recognition of face for follow-up link does not meet then delete target.
In above-mentioned first way, since in engineering practice, scene locating for Face datection related application exists
The features such as diversity, complexity, such as outdoor night and the mixed and disorderly environment of other environment light, this is based on face to using colour of skin etc.
The screening strategy of the colour space causes very big restriction.Because the face to be captured is by environment under these special screnes
The influence of light is very big, and face complexion is caused not to be able to satisfy the screening conditions based on color space being previously set.In addition it is exactly,
Strategy based on color filter is not applied for gray level image, scope of application relative narrower.
And for above-mentioned second method, similarly, in engineering practice, ring locating for the fore device of face detection system
The features such as diversity existing for border, complexity, extracts profile to the target captured or crucial point process causes very
Big difficulty.In addition, in actual operation, the pedestrian movement's randomness to be captured is very big, it is easy to occur rotary head, bow,
Phenomena such as blocking, the face captured all is greatly side face, along with the influence of environment light, even if that capture is people
Face target is also difficult accurately to extract its key point, these factors can largely reduce the accuracy rate of template matching.Meanwhile
Template matching method is more time-consuming, is less applicable for the real-time grasp shoot in engineering practice.
Based on the studies above, the embodiment of the invention provides a kind of schemes for reducing Face datection false detection rate, can utilize
The frame difference relationship between present frame and its previous frame image in image to be processed, and from multiple targets in current frame image
Moving Objects are filtered out in block, and environmental diversity interference caused by judgement is greatly reduced, reduces false detection rate.
Below in conjunction with attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Usually exist
The component of the embodiment of the present invention described and illustrated in attached drawing can be arranged and be designed with a variety of different configurations herein.Cause
This, is not intended to limit claimed invention to the detailed description of the embodiment of the present invention provided in the accompanying drawings below
Range, but it is merely representative of selected embodiment of the invention.Based on the embodiment of the present invention, those skilled in the art are not doing
Every other embodiment obtained under the premise of creative work out, shall fall within the protection scope of the present invention.
It should be noted that：Similar label and letter indicate similar terms in following attached drawing, therefore, once a certain Xiang Yi
It is defined in a attached drawing, does not then need that it is further defined and explained in subsequent attached drawing.Meanwhile of the invention
In description unless specifically defined or limited otherwise, term " installation ", " setting ", " connection " shall be understood in a broad sense, for example, can
To be to be fixedly connected, may be a detachable connection, or be integrally connected；It can be mechanical connection, be also possible to be electrically connected；It can
To be to be connected directly, the connection inside two elements can also be can be indirectly connected through an intermediary.For this field
For those of ordinary skill, the concrete meaning of above-mentioned term in the present invention can be understood with concrete condition.
Referring to Fig. 1, being the schematic block diagram of electronic equipment 100 provided in an embodiment of the present invention.In the present embodiment
In, the electronic equipment 100 can have the electronic equipment 100 of image collecting function for picture pick-up device, photographing device etc..Such as figure
Shown in 1, the electronic equipment 100 may include memory 130, processor 120 and be stored on the memory 130 and can
With the computer program run on the processor 120, the processor 120 sets the electronics when executing described program
Standby 100 methods for realizing reduction Face datection false detection rate of the invention.
It is directly or indirectly electrically connected between each other between the memory 130 and the processor 120, to realize number
According to transmission or interaction.For example, these elements can be realized electrically between each other by one or more communication bus or signal wire
Connection.The software being stored in the memory 130 in the form of software or firmware (Firmware) is stored in memory 130
Functional module, the software program and module that the processor 120 is stored in memory 130 by operation, such as present invention are real
The device 110 for applying the reduction Face datection false detection rate in example realizes this thereby executing various function application and data processing
The method of reduction Face datection false detection rate in inventive embodiments.
It is appreciated that structure shown in FIG. 1 is only to illustrate, the electronic equipment 100 may also include more than shown in Fig. 1
Perhaps less component or with the configuration different from shown in Fig. 1.Each component shown in Fig. 1 can use hardware, software
Or combinations thereof realize.
Referring to figure 2., Fig. 2 is a kind of side of reduction Face datection false detection rate applied to electronic equipment 100 shown in FIG. 1
The flow chart of method, below by the method includes each step be described in detail.
Step S101 obtains the previous frame image of current frame image to be detected and the current frame image.
Step S103 obtains multiple first object blocks from the current frame image, and obtains the previous frame image
In the second target block with each first object block respective coordinates.
The method that Face datection false detection rate is reduced provided by the present embodiment is applied after system captures link and face
Before the application such as identification.In view of in engineering practice, there is the pedestrian to be captured of the system for face snap movement to belong to
Property, so in the present embodiment, the Filtering system of moving target is devised first, to obtain movement pair from the image captured
As.
In the present embodiment, current frame image and the previous frame figure of the current frame image are obtained from the image captured
Picture.Wherein, the current frame image and previous frame image carried out pretreatment in advance, in the present embodiment, can be to present frame figure
Picture and previous frame image are pre-processed using neural network algorithm, multiple to retain in current frame image and previous frame image
There may be the regions of facial image.It can refer to prior art acquisition about pretreatment is carried out to image using neural network algorithm
More relevant knowledges, the present embodiment do not repeat.
But since the factors such as environment, real-time are influenced caused by preprocessing process, the region that is obtained after pretreatment
Generally there are erroneous detections, and treated, and region may be the region comprising face, it is also possible to include others, such as static object
Region.Therefore, the Moving Objects for being characterized as face object need to therefrom be detected using the frame difference strategy provided in the present embodiment,
To reduce false detection rate.
In the present embodiment, multiple first object blocks, the first object are obtained from pretreated current frame image
Block is the region that above-mentioned possibility includes facial image.Wherein, each first object block is in the current frame image
In have respective coordinate value, the coordinate value can be two-dimensional coordinate value, can each first object block central point seat
Whole coordinate value of the scale value as each first object block.Optionally, and from the previous frame image of the current frame image it obtains
There is the block of same coordinate value with each first object block, using as opposite with each first object block respectively
The second target block answered.It is subsequent can be according to the frame between each first object block and the second target block of corresponding coordinate
Poor relationship is to judge that each first object block is Moving Objects still for static object.
It in the present embodiment, can also be to the current frame image and the previous frame figure in order to reduce the treating capacity of image
As being scaled in proportion, by the current frame image and the previous frame image down to size appropriate.After carrying out again
Continuous processing so as to improve treatment effeciency, saves the processing time.
Step S105, according to the frame difference relationship of the second target block of each first object block and corresponding coordinate
Judge whether each first object block is static object.
Referring to Fig. 3, in the present embodiment, step S105 may include step S1051, step S1053 and step
Tri- sub-steps of S1055.
Step S1051 carries out difference fortune to the second target block of each first object block and corresponding coordinate
It calculates, to obtain the frame difference figure of each first object block and the second corresponding target block.
Step S1053, for each frame difference figure, by the pixel value of each pixel in the frame difference figure and the first default threshold
Value is compared, and obtains the corresponding frame difference equivalence of each pixel according to comparison result, equivalent according to the frame of each pixel difference
It obtains the frame difference and schemes frame difference total value between corresponding first object block and the second target block.
It in the present embodiment, can be only to progress size in order to avoid the frame difference bring time overhead of whole figure of calculating
Each second target block in each first object block and the previous frame image in current frame image after diminution carries out frame
Difference calculates.It optionally, can be to each first object block and corresponding coordinate for each first object block
Second target block carries out calculus of differences, to obtain the frame of the second target block of each first object and corresponding coordinate
Difference figure.
For each frame difference figure, the pixel value of each pixel in the frame difference figure can be obtained, and each pixel is corresponding
Pixel value is compared with the first preset threshold, and the frame difference figure pair is judged according to the comparison result of each pixel in the frame difference figure
Whether the first object block answered is static object.Wherein, first preset threshold may be set to 10, or other suitable
Suitable numerical value.
Referring to Fig. 4, in the present embodiment, step S1053 may include step S10531, step S10533, step
S10535, step S10537 and five sub-steps of step S10539.
The frame difference figure is divided into multiple sub-blocks by step S10531.
Step S10533, for each sub-block, by the pixel value of pixel each in the sub-block respectively with it is described
First preset threshold is compared.
The frame difference equivalence that pixel value is greater than the pixel of first preset threshold is set to 1, by pixel by step S10535
The frame difference equivalence that value is less than or equal to the pixel of first preset threshold is set to 0.
Step S10537 counts the equivalent summation of the frame difference for the pixel for including in the sub-block.
Step S10539, the summation equivalent to each sub-block corresponding frame difference adds up, to obtain described first
Frame difference total value between target block and second target block.
In the present embodiment, when whether judge each first object block is static object, in order to improve the accurate of judgement
Property, block division can be carried out to the corresponding frame difference figure of each first object block, further according to the pixel value of the pixel in each block
Situation is judged.Optionally, for each frame difference figure, the frame difference figure can be divided into multiple sub-blocks, such as drawn
It is divided into 16 sub-blocks of 4*4, naturally it is also possible to be divided into other quantity sub-block, not limit specifically this present embodiment
System.Only when carrying out the division of sub-block, it should be noted that in order to which subsequent processing is convenient, each sub-block should at least be protected
Card has multiple pixels, such as 2 or 3 etc..It is directed to each sub-block again, calculates separately it and detects difference.
Optionally, in the present embodiment, for each sub-block, the multiple pixels for including in the sub-block are obtained, and
The corresponding pixel value of each pixel is compared with first preset threshold respectively.It is counted for convenience in practical operation,
The frame difference equivalence of pixel that pixel value is greater than first preset threshold can be set to 1, pixel value is less than or equal to described the
The frame difference equivalence of the pixel of one preset threshold is set to 0.Frame difference equivalence herein can be equivalent to the frame difference of the pixel.Such as
This, it is pre- to be greater than described first for pixel value in the sum of frame difference equivalence of the pixel in the sub-block counted on as sub-block
If the number of the pixel of threshold value.
After carrying out obtaining the summation of frame difference equivalence of each sub-block respectively, such as can be equivalent by the frame difference of each sub-block
Summation be denoted as W respectivelyn,1, Wn,2..., Wn,16, wherein n indicates the number of the corresponding frame difference figure of each sub-block.It can be by each sub-district
The summation that the frame difference of block is equivalent adds up, to obtain the frame difference total value of the corresponding frame difference figure of each sub-block, i.e. the frame difference figure pair
Frame difference total value between the first object block answered and the second target block, is denoted as Wsum。
Step S1055 judges whether the first object block is static object according to the frame difference total value.
Referring to Fig. 5, in the present embodiment, step S1055 may include step S10551, step S10552, step
S10553, step S10554 and five sub-steps of step S10555.
It in the present embodiment, can basis when whether each first object block is static object in judging current frame image
Frame difference between the first object block and the second corresponding target block equivalent situation is judged, can also be tied
Close frame difference between the first object block and the second corresponding target block equivalent and the first object block and with
The case where the sum of frame difference equivalence of each sub-block in frame difference figure between its corresponding second target block, carrys out comprehensive descision.
Whether step S10551 detects the frame difference total value less than the second preset threshold, if being less than the described second default threshold
Value, then execute following steps S10552, if it is greater than or equal to second preset threshold, then execute following steps S10553.
Step S10552 determines that the first object block is static object.
Step S10553 chooses the equivalent maximum sub-district of summation of the frame difference of predetermined number from the multiple sub-block
Block.
Step S10554, the equivalent summation of the frame difference to the sub-block of selection add up, the accumulated value being calculated with
Ratio between the frame difference total value.
Step S10555, detects whether the ratio is greater than third predetermined threshold value, if more than the third predetermined threshold value, then
Execute step S10552.
In the present embodiment, for each first object block, the first object block and corresponding is detected
Whether detect poor equivalence less than the second preset threshold between second target block, wherein second preset threshold may be configured as 30 or
Other numerical value of person, it is as follows：
Wsum<30
If being less than second preset threshold, show between the first object block and corresponding second target block
It is not much different, it is most likely that the corresponding object of first object block is static object.On the contrary, if the first object block and
Frame difference total value between the second corresponding target block is more than or equal to second preset threshold, shows the first object area
Block may be dynamic object, it is contemplated that in this case, some other objects accidentally may be classified as dynamic object, example
Such as include the block of ground image for one, this may be made by the dynamic image generated when the ground because of pedestrian
The frame difference total value of the block of ground image is greater than second preset threshold, and then is dynamic object by the ground image erroneous detection.
It therefore, can be to the dynamic sub-block in the first object block when its frame difference total value is more than or equal to second preset threshold
The uniformity of distribution account for, to further decrease false detection rate.
Optionally, in the present embodiment, between the first object block and the second corresponding target block
When detecing poor total value more than or equal to second preset threshold, marked off from the corresponding frame difference figure of the first object block more
The equivalent maximum sub-block of summation of its frame difference is selected in a sub-block, for example, can select four sub-blocks or its
His quantity sub-block, to this, this embodiment is not specifically limited.
The equivalent summation of the frame of the sub-block selected difference is added up, and by cumulative obtained result and described first
Target block is corresponding to detect poor total value and makees ratio and obtain ratio value between the two.And by obtained ratio value and third predetermined threshold value
Carry out ratio, such as the third predetermined threshold value can be 0.75, or may be other numerical value.If obtained ratio value is greater than institute
Third predetermined threshold value is stated, that is, when meeting following formula, then can determine that the corresponding object of first object block is static object.
Wherein, WsumFrame difference total value between first object block and the second corresponding target block, Wmax1、
Wmax2、Wmax3、Wmax4Maximum first four pieces of summation of poor equivalence are detectd in the corresponding frame difference figure of respectively described first object block
The summation for detecing poor equivalence of sub-block.
Step S107 deletes the first object block for being determined as static object from the current frame image.
It should be noted that the face extremely normally captured individually, because not moving substantially, corresponding frame difference total value is because full
Foot is less than second preset threshold and is deleted, it is contemplated that capturing system is continuously to capture, so such situation is through reality
Trampling discovery substantially will not impact whole candid photograph rate.
In the present embodiment, the problem of poor strategy largely solves static erroneous detection is detectd by above, but it is right
The filtration result of dynamic erroneous detection is not ideal enough, therefore on the basis of the above, the present embodiment has also been devised based on deep learning side
The classifier of formula training from dynamic object to filter out face object.Referring to Fig. 6, reduction face inspection provided in this embodiment
The method for surveying false detection rate is further comprising the steps of：
Remaining first object block in current frame image after progress delete processing is input to foundation by step S109
In face classification device, to filter out face object from the remaining first object block.
Optionally, after through above step, static object can be detected from the image captured, and will be determined as quiet
The first object block of state object is deleted, and is interfered to avoid to subsequent processing.Remaining first object block after deleting
It is input in the face classification device of foundation, to filter out face object from remaining first object block.It should be noted that
Remaining first object block may be a first object block after being deleted, it is also possible to be multiple first object areas
Block is not specifically limited this in the present embodiment.
Referring to Fig. 7, in the present embodiment, the face classification device can be established by following steps：
Step S201 constructs the classifier network architecture based on convolutional neural networks.
Step S203, the multiple facial image positive samples that will acquire and multiple negative samples are separately input into the classifier net
To be trained to the facial image positive sample and the negative sample in network framework, to obtain the face classification device.
In the present embodiment, the classifier network architecture based on convolutional neural networks is constructed, point constructed in the present embodiment
The class device network architecture includes that three convolutional layers, four pond layers and two full articulamentums, specific structure are as shown in Figure 8.Its
In, convolutional layer one, convolutional layer two, convolutional layer three and full articulamentum one, full articulamentum two convolution kernel number be followed successively by respectively
8,16,32,32,2.Wherein, for convolution kernel size other than full articulamentum two is 1*1, the size of remaining several layers of convolution kernel is equal
For 3*3.Also, the step-length of above layers is 1.Wherein, in present networks framework, pond layer one, pond layer two, pond layer three
And the core size of pond layer four is 2*2, step-length is 2.
In the present embodiment, in order to make convergence faster, present networks framework active mode is all made of ReLU function, in addition, being
Over-fitting is prevented, in the last one down-sampling layer, i.e. pond layer four, is added to dropout mechanism, certain hidden layers is allowed to be weighed at random
Weight does not work.
In the present embodiment, after building the above-mentioned classifier network architecture, multiple facial images of training will be used for just
Sample and multiple negative samples, i.e., non-face image pattern, are separately input into the classifier network architecture of above-mentioned foundation, at this
The multiple facial image positive sample and the multiple negative sample are trained in the network architecture, to obtain above-mentioned face
Classifier, to make the subsequent judgment criteria to facial image.
Referring to Fig. 9, in the present embodiment, step S109 may include step S1091, step S1093 and step
Tri- sub-steps of S1095.
First object block remaining after deletion is input to the face classification device of foundation by step S1091.
Step S1093 detects the positive sample of facial image in the first object block and the face classification device after training
Between negative sample in the first degree of fitting and the first object block and the face classification device between this after training
Second degree of fitting.
Corresponding first degree of fitting of the first object block is compared, if institute by step S1095 with the second degree of fitting
The first degree of fitting is stated greater than second degree of fitting, then determines the first object block for face object.
In the present embodiment, remaining multiple first object blocks after above-mentioned deletion static object are input to foundation
In face classification device, to classify in the face classifier to each first object block.
In the present embodiment, detect each first object block for being input in the face classification device respectively with the face classification
The first degree of fitting between facial image positive sample and each first object block in device after training is respectively and by instructing
The second degree of fitting between negative sample after white silk, wherein the first degree of fitting and the second degree of fitting are small less than 1 and greater than 0
Number, and the first degree of fitting and the second degree of fitting and be 1.Detect whether corresponding first degree of fitting of each first object block is greater than
Second degree of fitting, if more than, then show that the first object block is higher to the degree of fitting of facial image positive sample, can determine that this
The corresponding object of one target block is facial image.Also, it is subsequent to will confirm that the first object block for facial image is reported and submitted
Otherwise face application link can be confirmed as erroneous detection object, carry out delete processing.So far, face screening link terminates.
Referring to Fig. 10, being missed for the reduction Face datection provided in an embodiment of the present invention applied to above-mentioned electronic equipment 100
The functional block diagram of the device 110 of inspection rate.Described device include image collection module 111, target block obtain module 112,
Judgment module 113, removing module 114 and screening module 115.
Described image obtains module 111 for obtaining upper the one of current frame image to be detected and the current frame image
Frame image.Described image, which obtains module 111, can be used for executing step S101 shown in Fig. 2, and specific operating method can refer to
The detailed description of step S101.
The target block obtains module 112 for obtaining multiple first object blocks from the current frame image, and
Obtain the second target block in the previous frame image with each first object block respective coordinates.The target block obtains
Modulus block 112 can be used for executing step S103 shown in Fig. 2, and specific operating method can refer to retouching in detail for step S103
It states.
The judgment module 113 is used for the second target block according to each first object block and corresponding coordinate
Frame difference relationship judge whether each first object block is static object.The judgment module 113 can be used for executing in Fig. 2
Shown step S105, specific operating method can refer to the detailed description of step S105.
The removing module 114 is used for when being determined as static object, will be determined as the first object block of static object
It is deleted from the current frame image.The removing module 114 can be used for executing step S107 shown in Fig. 2, specific to grasp
It can refer to the detailed description of step S107 as method.
The screening module 115 is defeated for remaining first object block in the current frame image after carrying out delete processing
Enter into the face classification device of foundation, to filter out face object from the remaining first object block.The screening module
115 can be used for executing step S109 shown in Fig. 6, and specific operating method can refer to the detailed description of step S109.
It is apparent to those skilled in the art that for convenience and simplicity of description, the device of foregoing description
Specific work process, no longer can excessively be repeated herein with reference to the corresponding process in preceding method.
Through the above description of the embodiments, those skilled in the art can be understood that the present invention can lead to
Hardware realization is crossed, the mode of necessary general hardware platform can also be added to realize by software.Based on this understanding, of the invention
Technical solution can be embodied in the form of software products, which can store is situated between in non-volatile memories
In matter (can be CD-ROM, USB flash disk, mobile hard disk etc.), including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) execute method described in each implement scene of the present invention.
In conclusion the embodiment of the present invention is provided in the method, apparatus and electronic equipment for reducing Face datection false detection rate
100, it is obtained by obtaining multiple first object blocks from current frame image, and from the previous frame image of the current frame image
There are multiple second target blocks of respective coordinates with each first object block, according to each first object block and corresponding
The frame difference relationship of second target block is to judge whether each first object block is static object, and when being determined as static object
Corresponding first object block is deleted.On this basis, remaining first object block is input to the face classification of foundation
In device, to filter out facial image from remaining first object block.By above step, can accurately be sentenced using frame difference relationship
Moving Objects in the disconnected target captured out, reduce environmental diversity interference caused by detection, improve the standard of judgement
True property.Further, facial image is filtered out from Moving Objects in conjunction with the face classification device of deep learning, further filter out mistake
Inspection.
In embodiment provided herein, it should be understood that disclosed device and method, it can also be by other
Mode realize.The apparatus embodiments described above are merely exemplary, for example, the flow chart and block diagram in attached drawing are shown
Architectural framework in the cards, function and the behaviour of devices in accordance with embodiments of the present invention, method and computer program product
Make.In this regard, each box in flowchart or block diagram can represent a part of a module, section or code, institute
The a part for stating module, section or code includes one or more executable instructions for implementing the specified logical function.
It should also be noted that function marked in the box can also be to be different from attached drawing in some implementations as replacement
The sequence marked occurs.For example, two continuous boxes can actually be basically executed in parallel, they sometimes can also be by
Opposite sequence executes, and this depends on the function involved.It is also noted that each box in block diagram and or flow chart,
And the combination of the box in block diagram and or flow chart, hardware can be based on the defined function of execution or the dedicated of movement
System realize, or can realize using a combination of dedicated hardware and computer instructions.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or equipment for including a series of elements not only includes those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or equipment institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including institute
State in the process, method, article or equipment of element that there is also other identical elements.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.It should be noted that：Similar label and letter exist
Similar terms are indicated in following attached drawing, therefore, once being defined in a certain Xiang Yi attached drawing, are then not required in subsequent attached drawing
It is further defined and explained.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (10)
1. a kind of method for reducing Face datection false detection rate, which is characterized in that the method includes：
Obtain the previous frame image of current frame image to be detected and the current frame image；
Multiple first object blocks are obtained from the current frame image, and are obtained in the previous frame image with each described first
Second target block of target block respective coordinates；
Each described is judged according to the frame difference relationship of the second target block of each first object block and corresponding coordinate
Whether one target block is static object；
If it is determined that then the first object block for being determined as static object is deleted from the current frame image for static dynamic.
2. the method according to claim 1 for reducing Face datection false detection rate, which is characterized in that described to be determined as static state
The first object block of object from the current frame image delete the step of after, the method also includes：
Remaining first object block in current frame image after progress delete processing is input in the face classification device of foundation,
To filter out face object from the remaining first object block.
3. the method according to claim 1 for reducing Face datection false detection rate, which is characterized in that described according to each described the
The frame difference relationship of one target block and the second target block of corresponding coordinate judge each first object block whether be
The step of static object, including：
Calculus of differences is carried out to the second target block of each first object block and corresponding coordinate, it is each described to obtain
The frame difference figure of first object block and the second corresponding target block；
For each frame difference figure, the pixel value of each pixel in the frame difference figure is compared with the first preset threshold, root
It is equivalent that the corresponding frame difference of each pixel is obtained according to comparison result；
The corresponding first object block of the frame difference figure and the second target block are obtained according to the frame difference equivalence of each pixel
Between frame difference total value；
Judge whether the first object block is static object according to the frame difference total value.
4. the method according to claim 3 for reducing Face datection false detection rate, which is characterized in that described by the frame difference figure
In the pixel value of each pixel be compared with the first preset threshold, corresponding frame difference of each pixel etc. is obtained according to comparison result
Value, according to the frame difference equivalence of each pixel obtain the corresponding first object block of the frame difference figure and the second target block it
Between frame difference total value the step of, including：
The frame difference figure is divided into multiple sub-blocks；
For each sub-block, the pixel value of pixel each in the sub-block is carried out with first preset threshold respectively
Compare；
The frame difference equivalence that pixel value is greater than the pixel of first preset threshold is set to 1, pixel value is less than or equal to described
The frame difference equivalence of the pixel of first preset threshold is set to 0；
Count the equivalent summation of the frame difference for the pixel for including in the sub-block；
The summation equivalent to each sub-block corresponding frame difference adds up, to obtain the first object block and described the
Frame difference total value between two target blocks.
5. the method according to claim 4 for reducing Face datection false detection rate, which is characterized in that described poor according to the frame
Total value judges the step of whether the first object block is static object, including：
The frame difference total value is detected whether less than the second preset threshold, if being less than second preset threshold, determines described the
One target block is static object；
If it is greater than or equal to second preset threshold, then the total of the frame difference equivalence of predetermined number is chosen from the multiple sub-block
With maximum sub-block；
The summation equivalent to the frame difference of the sub-block of selection adds up, between the accumulated value being calculated and the frame difference total value
Ratio；
It detects whether the ratio is greater than third predetermined threshold value, if more than the third predetermined threshold value, then determines first mesh
Mark block is static object.
6. the method according to claim 2 for reducing Face datection false detection rate, which is characterized in that the face classification device is logical
Cross following steps acquisition：
Construct the classifier network architecture based on convolutional neural networks；
The multiple facial image positive samples that will acquire and multiple negative samples are separately input into the classifier network architecture with right
The facial image positive sample and the negative sample are trained, to obtain the face classification device.
7. the method according to claim 6 for reducing Face datection false detection rate, which is characterized in that described to carry out at deletion
Remaining first object block is input in the face classification device of foundation in current frame image after reason, with from this remaining first
The step of face object is filtered out in target block, including：
First object block remaining after deletion is input to the face classification device of foundation；
First detected between the facial image positive sample in the first object block and the face classification device after training is intended
The second degree of fitting between negative sample in the right and described first object block and the face classification device after training；
Corresponding first degree of fitting of the first object block is compared with the second degree of fitting, if first degree of fitting is big
In second degree of fitting, then determine the first object block for face object.
8. the method according to claim 1 for reducing Face datection false detection rate, which is characterized in that described from the present frame
Multiple first object blocks are obtained in image, and obtain in the previous frame image with each first object block respective coordinates
The second target block the step of before, the method also includes：
The current frame image and the previous frame image are scaled in proportion.
9. a kind of device for reducing Face datection false detection rate, which is characterized in that described device includes：
Image collection module, for obtaining the previous frame image of current frame image to be detected and the current frame image；
Target block obtains module, for obtaining multiple first object blocks from the current frame image, and obtains on described
In one frame image with the second target block of each first object block respective coordinates；
Judgment module, for the frame difference relationship according to each first object block and the second target block of corresponding coordinate
Judge whether each first object block is static object；
Removing module will be determined as the first object block of static object from described current for when being determined as static object
It is deleted in frame image.
10. a kind of electronic equipment, which is characterized in that including：
Memory；
Processor；And
The device of Face datection false detection rate is reduced, including one or more is stored in the memory and is held by the processor
Capable software function module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711462162.7A CN108932465B (en) | 2017-12-28 | 2017-12-28 | Method and device for reducing false detection rate of face detection and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711462162.7A CN108932465B (en) | 2017-12-28 | 2017-12-28 | Method and device for reducing false detection rate of face detection and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108932465A true CN108932465A (en) | 2018-12-04 |
| CN108932465B CN108932465B (en) | 2021-02-02 |
Family
ID=64448331
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711462162.7A Active CN108932465B (en) | 2017-12-28 | 2017-12-28 | Method and device for reducing false detection rate of face detection and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108932465B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414444A (en) * | 2019-07-31 | 2019-11-05 | 中国工商银行股份有限公司 | Face identification method and device |
| CN111528652A (en) * | 2020-07-09 | 2020-08-14 | 北京每日优鲜电子商务有限公司 | Method and device for identifying commodities in intelligent container |
| CN113554008A (en) * | 2021-09-18 | 2021-10-26 | 深圳市安软慧视科技有限公司 | Method and device for detecting static object in area, electronic equipment and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102129572A (en) * | 2011-02-25 | 2011-07-20 | 杭州海康威视软件有限公司 | Face detection method and device adopting cascade classifier |
| CN102855496A (en) * | 2012-08-24 | 2013-01-02 | 苏州大学 | Method and system for authenticating shielded face |
| CN103577838A (en) * | 2013-11-25 | 2014-02-12 | 苏州大学 | Face recognition method and device |
| CN104616006A (en) * | 2015-03-11 | 2015-05-13 | 湖南智慧平安科技有限公司 | Surveillance video oriented bearded face detection 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 |
| US20170069104A1 (en) * | 2015-09-04 | 2017-03-09 | Canon Kabushiki Kaisha | Image processing apparatus and image processing method |
-
2017
- 2017-12-28 CN CN201711462162.7A patent/CN108932465B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102129572A (en) * | 2011-02-25 | 2011-07-20 | 杭州海康威视软件有限公司 | Face detection method and device adopting cascade classifier |
| CN102855496A (en) * | 2012-08-24 | 2013-01-02 | 苏州大学 | Method and system for authenticating shielded face |
| CN103577838A (en) * | 2013-11-25 | 2014-02-12 | 苏州大学 | Face recognition method and device |
| CN104616006A (en) * | 2015-03-11 | 2015-05-13 | 湖南智慧平安科技有限公司 | Surveillance video oriented bearded face detection 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 |
| US20170069104A1 (en) * | 2015-09-04 | 2017-03-09 | Canon Kabushiki Kaisha | Image processing apparatus and image processing method |
Non-Patent Citations (1)
| Title |
|---|
| 胡建华: "基于分块差分与背景减除的运动检测", 《电子测量技术》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414444A (en) * | 2019-07-31 | 2019-11-05 | 中国工商银行股份有限公司 | Face identification method and device |
| CN111528652A (en) * | 2020-07-09 | 2020-08-14 | 北京每日优鲜电子商务有限公司 | Method and device for identifying commodities in intelligent container |
| CN111528652B (en) * | 2020-07-09 | 2021-11-09 | 北京每日优鲜电子商务有限公司 | Method and device for identifying commodities in intelligent container |
| CN113554008A (en) * | 2021-09-18 | 2021-10-26 | 深圳市安软慧视科技有限公司 | Method and device for detecting static object in area, electronic equipment and storage medium |
| CN113554008B (en) * | 2021-09-18 | 2021-12-31 | 深圳市安软慧视科技有限公司 | Method and device for detecting static object in area, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108932465B (en) | 2021-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108446617B (en) | Side face interference resistant rapid human face detection method | |
| CN106960195B (en) | Crowd counting method and device based on deep learning | |
| CN110070033B (en) | Method for detecting wearing state of safety helmet in dangerous working area in power field | |
| CN107527009B (en) | Remnant detection method based on YOLO target detection | |
| CN107123131B (en) | Moving target detection method based on deep learning | |
| CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
| CN101271514B (en) | Image detection method and device for fast object detection and objective output | |
| CN109376667A (en) | Object detection method, device and electronic equipment | |
| CN106951870B (en) | Intelligent detection and early warning method for active visual attention of significant events of surveillance video | |
| CN109145708B (en) | Pedestrian flow statistical method based on RGB and D information fusion | |
| CN105184818B (en) | A kind of video monitoring anomaly detection method and its detecting system | |
| CN108416250A (en) | Demographic method and device | |
| CN107688784A (en) | A kind of character identifying method and storage medium based on further feature and shallow-layer Fusion Features | |
| CN103092930B (en) | Method of generation of video abstract and device of generation of video abstract | |
| CN105930822A (en) | Human face snapshot method and system | |
| CN106778645A (en) | A kind of image processing method and device | |
| CN109241896A (en) | A kind of channel security detection method, device and electronic equipment | |
| CN109766779A (en) | It hovers personal identification method and Related product | |
| CN108010025A (en) | Switch and indicator lamp positioning and state identification method of screen cabinet based on RCNN | |
| US10282601B2 (en) | Electronic device and gesture recognition method applied therein | |
| CN108932465A (en) | Reduce the method, apparatus and electronic equipment of Face datection false detection rate | |
| CN108090406A (en) | Face identification method and system | |
| CN109492647A (en) | A kind of power grid robot barrier object recognition methods | |
| CN111445442B (en) | Crowd counting method and device based on neural network, server and storage medium | |
| CN113688820A (en) | Stroboscopic stripe information identification method and device and electronic equipment |
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 |