CN105701457A - DC electromagnetic relay device based on face identification control and control method thereof - Google Patents

Abstract

The present invention discloses a DC electromagnetic relay device based on face identification control. The DC electromagnetic relay device based on face identification control comprises a face identification module, a power supply switch controller, a voice broadcasting device, a DC electromagnetic relay and a liquid crystal display, the face identification module is connected with the liquid crystal display, the power supply switch controller and the voice broadcasting device separately, and the power supply switch controller is connected with a low-voltage control circuit of the DC electromagnetic relay. According to the present invention, by a method of identifying the real-time face images of the authorized users, the purpose of controlling the on-off of a DC electromagnetic relay circuit is achieved, and the electrical equipment can be controlled simply, rapidly, safely and intelligently. The present invention also discloses a control method of the DC electromagnetic relay device based on face identification control. The method comprises the concrete steps of 1 carrying out the pre-processing on the acquired face images; 2 carrying out the pre-processing on the feature face vectors outputted by a strong classifier and then utilizing an Angle Contex matrix to carry out the similar face matching on the feature vectors.

Description

The direct current electromagnetic relay device controlled based on recognition of face and control method thereof

Technical field

The invention belongs to electric control system technical field, relate to a kind of direct current electromagnetic relay device controlled based on recognition of face, the invention still further relates to the control method of this direct current electromagnetic relay device。

Background technology

Along with information technology permeates to service for life each side, industry internet, cloud computing, continuous fusion application between big data and new technology of Internet, the rise adding Internet of Things promotes industry to move towards intelligent, and the demand of electrical equipment is also progressively tended to intelligentized side's exhibition by power consumer。Traditional electromagnetic relay is commonly used in automatic control circuit, uses less electric current and Control of Voltage to have the electrical equipment of larger current and high voltage。If being used for controlling the special equipment of electric aspect or high-risk equipment by traditional electromagnetic relay switch, it is easy to cause the illegal operation of layman, cause the infringement to electrical equipment, it could even be possible to make whole electrical system paralyse。Traditional electromagnetic relay switch is for controlling special equipment or the safe and reliable problem of high-risk equipment existence。

Summary of the invention

It is an object of the invention to provide a kind of direct current electromagnetic relay device controlled based on recognition of face, can by identifying that the method for authorized user's real-time face image reaches to control the purpose of the break-make of direct current electromagnetic relay circuit and control electrical equipment that can be simple, quick, safe and reliable, intelligent。

The technical solution used in the present invention is, a kind of direct current electromagnetic relay device controlled based on recognition of face, including face recognition module, power supply switch controller, voice announcer, direct current electromagnetic relay, liquid crystal display, face recognition module respectively with liquid crystal display, on and off switch controls module and voice announcer connects, on and off switch controls module and is connected with the low-voltage control circuit of direct current electromagnetic relay, the facial image of user can be demonstrated on a liquid crystal display in real time when face recognition module recognizes user's face, if on and off switch control module can export the low-voltage control circuit of 24V DC voltage supply direct current electromagnetic relay after face recognition module identification success。

It is another object of the invention to provide the control method of the above-mentioned direct current electromagnetic relay device controlled based on recognition of face。

Another technical scheme of the present invention is, based on the control method of the direct current electromagnetic relay device that recognition of face controls, specifically implements according to following steps:

Step 1, carries out pretreatment to the facial image collected, and adopts Haar wavelet basis function to calculate rectangular characteristic the structural classification device of image, face is detected,

Step 2, after eigenface vector R (D) pretreatment that strong classifier is exported, characteristic vector R (D) is carried out similar face coupling by recycling AngleContex matrix, improve recognition of face precision, set up the face database of authorized user in advance, then the face characteristic extracted utilize AngleContex matrix be calculated it and the similarity authorizing face database set up, it may be judged whether for authorized user's face。

The feature of the present invention also resides in,

Comprising the concrete steps that of step 1:

Step 1.1, rectangular characteristic value is all grey scale pixel value sums and the difference of all grey scale pixel value sums in white portion in black region, have employed a kind of rectangular characteristic being similar to Haar wavelet basis function, rectangular characteristic value in detection window, rectangular characteristic value can be calculated by the gray value of integral image pixel, in integral image, coordinate is that the gray value of (X, Y) pixel is equal to all grey scale pixel values in correspondence position upper left side cumulative on original image；

The computing formula of integral image is as follows:

$I(X,Y)=\underset{x\≤X}{\Σ}\underset{y\≤Y}{\Σ}i(x,y)---(1-1)$

I (X, Y) is the image after integration, i (x, y) for original image,

Step 1.2, trying to achieve the integral image I (X, Y) of pixel as a sequence, forms the integrogram matrix of this figure, finds the set D of matrix exgenvalue according to integrogram matrix,

D={a₁,a₂,a₃,a₄,........a_m}(1-2)

Step 1.3, in rectangular window, builds multistage classifier and obtains the classification function of optimal threshold, matrix exgenvalue is carried out best classification,

Step 1.4, utilizes binarization of gray value and integral projection method to carry out eye location in the human face region detected；With eyes coordinates for datum mark, two eyes are made to be in same level line by translating, rotating, by scale transformation, facial image is carried out size to unify, facial image is carried out grayscale normalization, average and the mean square deviation of facial contour inner region gray scale is calculated by histogram equalization, image smoothing, gray scale normalization etc., view picture facial image is carried out greyscale transformation, highlights face key facial features。

Concretely comprising the following steps of step 1.3:

Step 1.3.1, constructs training sample set L:

L={ (D₁,K₁),(D₂,K₂),......(D_m,K_m)}(1-3)

Wherein D represents the matrix characteristic vector of sample, and K represents training result, and m is training total number of samples, m=1,2,3 ... .n；

Step 1.3.2, the initial probability distribution of sample L is sample weights:

N represents number of training, and the iterations of grader is t=1,2,3,4.......T, and T is any positive integer, calculates the weights of normalization sample:

$q_{t,m}=\frac{W_{t,m}}{\underset{j=1}{\Σ}{W}_{t,m}}---(1-4)$

Wherein j=1,2,3 ... .n；

Step 1.3.3, calls weak learning algorithm to the matrix characteristic vector D of each sample, trains a Weak Classifier H_j(D_m), and calculate the error rate ξ of this grader_j

$H_j\left(D_m\right)=\{\begin{array}{cc}1;& p_j{f}_j\≥p_j{\mathrm{\θ}}_j\\ 0;& otherwise\end{array}---(1-5)$

Wherein θ_jIt is the threshold value of Weak Classifier, p_jFor the biasing of grader, the direction of majorization inequality, f_jFor the eigenvalue of eigenmatrix,

${\mathrm{\ξ}}_j=\underset{m}{\Σ}{q}_{t,m}|H_j\left(D_m\right)-K_m|---(1-6)$

Step 1.3.4, selects have minimal error rate ξ_tCorresponding Weak Classifier H_t(D_m) as optimal classification device, namely

ξ_t=min ξ_j(1-7)

Step 1.3.5, readjusting sample weights according to optimal classification device is:

$W_{t+1,m}=W_{t,m}{{\mathrm{\β}}_t}^{1-e_m}---(1-8)$

Wherein m-th sample correct judgment then e_mOtherwise=0 e_m=1. so can increase grader error sample weight, reduce the weight of correct sample of classifying, and during next iteration, the selection of Weak Classifier will pay attention to the sample of last mistake classification error,

${\mathrm{\β}}_t=\frac{{\mathrm{\ξ}}_t}{1-{\mathrm{\ξ}}_t}---(1-9)$

β_tThe coefficient of sample weights is readjusted for optimal classification device；

Step 1.3.6, generates strong classifier:

$R\left(D\right)=\underset{t=1}{\overset{T}{\Σ}}log\frac{1}{{\mathrm{\β}}_t}{H}_t\left(D_m\right)---(1-10)$

IfThen grader output result is 1, then it is assumed that this detection window is likely human face region, and R (D) is the eigenface vector of output；

The multistage classifier utilizing structure can get rid of most of non-face region, detects almost possible human face region；If every first-level class device is output as 1, then it is assumed that this detection window is likely human face region, the window of detection is input to next stage grader and proceeds to judge, otherwise, in this level, detection window is excluded as non-face region。

Step 2 particularly as follows:

When extracting face characteristic, it is identical for belonging to the human face characteristic point distance with a part, and the difference of face is mainly manifested in the difference of face each several part contour line curvature, angle, can with 8 degree of divisions as face characteristic average distance, by 45 regions of plane division being zero with datum mark, generate AngleContex matrix, calculate the number of other discrete points falling into regional

Specifically implement according to following steps:

Step 2.1, on the eigenface that strong classifier obtains, each datum mark is selected for 9 character references points with nose, corners of the mouth midpoint, point, left eyebrow, left eye angle, the left corners of the mouth, right eyebrow, right eye angle, the right corners of the mouth, 45 regions (bin) are chosen with 8 degree of angles for dividing interval, thus can obtain 9*45 AngleContex matrix

Step 2.2, utilizes angular histogram to intersect method and the similarity of 9*45 the AngleContex matrix calculus face to be matched obtained and targeted customer's face, and computing formula is as follows:

$P(s,v)=\underset{l,g}{\Σ}min\[R_{l,g}\left(D\right),{R^{\′}}_{l,g}\left(D\right)\]---(2-1)$

$M(R,R^{\′})=\frac{P(s,v)}{9*45}---(2-2)$

P (s, v) is the crossing distance of two characteristic vectors, and M (R, R') is the ratio of 9*45 overall area (bin) pixel shared by each zoning (bin) in two rectangular histograms,

Wherein, R_l,gRepresent the AngleContex matrix of face to be matched, R'_l,gThe AngleContex matrix of expression targeted customer's face, l=1,2 ..., 9 represent 9 character references points；G=1,2,3,4,5,6 ..., 44,45 expressions choose 45 regions with 8 degree of angles for dividing interval；

Step 2.3, the value calculating M (R, R') between 0～1, can take M (R, R') maximum just for the similar face that matching degree is the highest,

Step 2.4, after when similar face, the match is successful, voice announcer 3 can report the voice of " identifying successfully ", and otherwise, voice announcer 3 can report the voice of " identifying mistake ",

Wherein human face similarity degree coupling, on the eigenface image obtained, with nose, corners of the mouth midpoint, point, left eyebrow, left eye angle, the left corners of the mouth, right eyebrow, right eye angle, the right corners of the mouth for 9 character references points, each datum mark selected on the eigenface image obtained through strong classifier, 45 regions (bin) are chosen with 8 degree of angles for dividing interval, thus can obtaining 9*45 AngleContex matrix, multizone divides and improves matching precision as far as possible。

The invention has the beneficial effects as follows, compared with existing system, implementing the equipment used in the present invention less, simple in construction, with low cost, feasibility and the safety of method are higher。The combination of face recognition technology and direct current electromagnetic relay can be made full use of, the effective special equipment controlled in electrical system or high-risk equipment, avoid the improper operation of disabled user, for industrial implementation intellectuality, there is long-range meaning, face recognition technology is adopted to use user to carry out identification electrical equipment, giving the control loop that user identity authority remotely controls direct current electromagnetic relay, controlling thus reaching authorized user's operation to special equipment or high-risk equipment。

Accompanying drawing explanation

Fig. 1 is the structural representation of a kind of direct current electromagnetic relay controlled based on recognition of face of the present invention；

Fig. 2 is a kind of based on the power on-off control circuit figure in the direct current electromagnetic relay of recognition of face control of the present invention；

Fig. 3 is a kind of flow chart based on the face recognition module recognition methods in the direct current electromagnetic relay of recognition of face control of the present invention。

In figure, 1. face recognition module, 2. power supply switch controller, 3. voice announcer, 4. direct current electromagnetic relay, 5. liquid crystal display。

Detailed description of the invention

Below in conjunction with accompanying drawing and specific embodiment, describe the present invention。

The invention provides a kind of direct current electromagnetic relay device controlled based on recognition of face, as it is shown in figure 1, include face recognition module 1, power supply switch controller 2, voice announcer 3, direct current electromagnetic relay 4, liquid crystal display 5。

Face recognition module 1 controls module 2 with liquid crystal display 5, on and off switch respectively and voice announcer 3 is connected, and on and off switch controls module 2 and is connected with the low-voltage control circuit of direct current electromagnetic relay 4。The facial image of user can be demonstrated on liquid crystal display 5 in real time, if face recognition module 1 identifies that after successfully, on and off switch control module 2 can export the low-voltage control circuit of 24V DC voltage supply direct current electromagnetic relay 4 when face recognition module 1 recognizes user's face。

As in figure 2 it is shown, recognition of face device utilizes camera collection facial image, it is possible to recognition of face device 1 is arranged on indoor, reach remotely control electromagnetic relay and gather video image purpose clearly。When user needs to open electrical equipment, before directly walking to the recognition of face device 1 of indoor, the facial information of active user is detected and positions by the photographic head on recognition of face device 1, uses user it will be clear that facial image on liquid crystal display 5。Utilize video capture circuit that the video signal of camera acquisition is converted into digital signal, and the digital picture after gathering is carried out subsequent treatment。

As shown in Figure 2, its work process is the power on-off control circuit figure of apparatus of the present invention, and after voice announcer 3 reports the voice of " identifying successfully ", power supply switch controller 2 can be exported high level that signal is 1 by recognition of face device 1,

Audion Q119013 in power supply switch controller will turn on, the grid so making the field effect transistor Q7 of enhancement mode P-channel is in negative potential, the source electrode of field effect transistor Q7 has the voltage of-24V under the DC voltage effect of 24V, operation principle according to field effect transistor Q7, make QX end, and export the low-voltage control circuit of 24V DC voltage conducting direct current electromagnetic relay 4, the armature of relay 4 is inhaled by electric magnet, after movable contact contacts with stationary contact simultaneously, authorized user can remote opening electrical equipment。Otherwise, voice announcer 3 reports the voice of " identifying mistake ", power supply switch controller 2 output low level, audion is in cut-off state, the grid of field effect transistor Q7 and source electrode both end voltage are 0, and field effect transistor Q7 is in cut-off state, and the low-voltage control circuit of direct current electromagnetic relay 4 will not be energized, armature remains static, and the electrical equipment of peripheral circuit fails normally-open。

A kind of control method of the direct current electromagnetic relay device controlled based on recognition of face, as it is shown on figure 3, specifically implement according to following steps:

Step 1, carries out pretreatment to the facial image collected, and adopts Haar wavelet basis function to calculate rectangular characteristic the structural classification device of image, face is detected, and concrete step is:

Step 1.1, rectangular characteristic value is all grey scale pixel value sums and the difference of all grey scale pixel value sums in white portion in black region, have employed a kind of rectangular characteristic being similar to Haar wavelet basis function, the rectangular characteristic value in detection window。Rectangular characteristic value can be calculated by the gray value of integral image pixel, and in integral image, coordinate is that the gray value of (X, Y) pixel is equal to all grey scale pixel values in correspondence position upper left side cumulative on original image；

The computing formula of integral image is as follows:

$I(X,Y)=\underset{x\≤X}{\Σ}\underset{y\≤Y}{\Σ}i(x,y)---(1-1)$

I (X, Y) is the image after integration, i (x, y) for original image,

Step 1.2, trying to achieve the integral image I (X, Y) of pixel as a sequence, forms the integrogram matrix of this figure, finds the set D of matrix exgenvalue according to integrogram matrix,

D={a₁,a₂,a₃,a₄,........a_m}(1-2)

Step 1.3, in rectangular window, builds multistage classifier and obtains the classification function of optimal threshold, matrix exgenvalue is carried out best classification, and specific algorithm is as follows:

Step 1.3.1, constructs training sample set L:

L={ (D₁,K₁),(D₂,K₂),......(D_m,K_m)}(1-3)

Wherein D represents the matrix characteristic vector of sample, and K represents training result, and m is training total number of samples, m=1,2,3 ... .n；

Step 1.3.2, the initial probability distribution of sample L is sample weights:

N represents number of training, and the iterations of grader is t=1,2,3,4.......T, and T is any positive integer, calculates the weights of normalization sample:

$q_{t,m}=\frac{W_{t,m}}{\underset{j=1}{\Σ}{W}_{t,m}}---(1-4)$

Wherein j=1,2,3 ... .n；

Step 1.3.3, calls weak learning algorithm to the matrix characteristic vector D of each sample, trains a Weak Classifier H_j(D_m), and calculate the error rate ξ of this grader_j

$H_j\left(D_m\right)=\{\begin{array}{cc}1;& p_j{f}_j\≥p_j{\mathrm{\θ}}_j\\ 0;& otherwise\end{array}---(1-5)$

Wherein θ_jIt is the threshold value of Weak Classifier, p_jFor the biasing of grader, the direction of majorization inequality, f_jEigenvalue for eigenmatrix。

${\mathrm{\ξ}}_j=\underset{m}{\Σ}{q}_{t,m}|H_j\left(D_m\right)-K_m|---(1-6)$

Step 1.3.4, selects have minimal error rate ξ_tCorresponding Weak Classifier H_t(D_m) as optimal classification device, namely

ξ_t=min ξ_j(1-7)

Step 1.3.5, readjusting sample weights according to optimal classification device is:

$W_{t+1,m}=W_{t,m}{{\mathrm{\β}}_t}^{1-e_m}---(1-8)$

Wherein m-th sample correct judgment then e_mOtherwise=0 e_m=1. so can increase grader error sample weight, reduce the weight of correct sample of classifying, and during next iteration, the selection of Weak Classifier will pay attention to the sample of last mistake classification error。

${\mathrm{\β}}_t=\frac{{\mathrm{\ξ}}_t}{1-{\mathrm{\ξ}}_t}---(1-9)$

β_tThe coefficient of sample weights is readjusted for optimal classification device；

Step 1.3.6, generates strong classifier:

$R\left(D\right)=\underset{t=1}{\overset{T}{\Σ}}log\frac{1}{{\mathrm{\β}}_t}{H}_t\left(D_m\right)---(1-10)$

IfThen grader output result is 1, then it is assumed that this detection window is likely human face region, and R (D) is the eigenface vector of output；

The multistage classifier utilizing structure can get rid of most of non-face region, detects almost possible human face region；If every first-level class device is output as 1, then it is assumed that this detection window is likely human face region, the window of detection is input to next stage grader and proceeds to judge。Otherwise, in this level, detection window is excluded as non-face region；

Step 1.4, utilizes binarization of gray value and integral projection method to carry out eye location in the human face region detected；With eyes coordinates for datum mark, two eyes are made to be in same level line by translating, rotating, by scale transformation, facial image is carried out size to unify, facial image is carried out grayscale normalization, average and the mean square deviation of facial contour inner region gray scale is calculated by histogram equalization, image smoothing, gray scale normalization etc., view picture facial image is carried out greyscale transformation, highlights face key facial features；

Step 2, after eigenface vector R (D) pretreatment that strong classifier is exported, characteristic vector R (D) is carried out similar face coupling by recycling AngleContex matrix, improves recognition of face precision。Set up the face database of authorized user in advance, then the face characteristic extracted utilize AngleContex matrix be calculated it and the similarity authorizing face database set up, it may be judged whether for authorized user's face, concrete step is:

When extracting face characteristic, it is identical for belonging to the human face characteristic point distance with a part, and the difference of face is mainly manifested in the difference of face each several part contour line curvature, angle, can with 8 degree of divisions as face characteristic average distance, by 45 regions of plane division being zero with datum mark, generate AngleContex matrix, calculate the number of other discrete points falling into regional

Specific algorithm illustrates:

Step 2.1, on the eigenface that strong classifier obtains, each datum mark is selected for 9 character references points with nose, corners of the mouth midpoint, point, left eyebrow, left eye angle, the left corners of the mouth, right eyebrow, right eye angle, the right corners of the mouth, choose 45 regions (bin) with 8 degree of angles for dividing interval, thus can obtain 9*45 AngleContex matrix；

Step 2.2, utilizes angular histogram to intersect method and the similarity of 9*45 the AngleContex matrix calculus face to be matched obtained and targeted customer's face, and computing formula is as follows:

$P(s,v)=\underset{l,g}{\Σ}min\[R_{l,g}\left(D\right),{R^{\′}}_{l,g}\left(D\right)\]---(2-1)$

$M(R,R^{\′})=\frac{P(s,v)}{9*45}---(2-2)$

(s, v) is the crossing distance of two characteristic vectors to P, and M (R, R') is the ratio of 9*45 overall area (bin) pixel shared by each zoning (bin) in two rectangular histograms；

Wherein, R_l,gRepresent the AngleContex matrix of face to be matched, R'_l,gThe AngleContex matrix of expression targeted customer's face, l=1,2 ..., 9 represent 9 character references points；G=1,2,3,4,5,6 ..., 44,45 expressions choose 45 regions with 8 degree of angles for dividing interval；

Step 2.3, the value calculating M (R, R') between 0～1, can take M (R, R') maximum just for similar face that matching degree is the highest；

Step 2.4, after when similar face, the match is successful, voice announcer 3 can report the voice of " identifying successfully ", and otherwise, voice announcer 3 can report the voice of " identifying mistake "。

Its feature of the control method of the present invention is mainly reflected in the following aspects:

1) setting up face database, the face characteristic of authorized user carries out registration typing, in order to improve the safety of face database, the data preserved in face database can only be checked by the user having permission and change。

2) use video camera that user is carried out face image collection, the image gathered is carried out Face datection and location, and can to given arbitrary image, it is determined that wherein whether there is face, and provide the information such as the position of face, size and state。

3) utilize video capture circuit that the video signal of camera acquisition is converted into digital signal, and the digital picture after gathering is carried out pretreatment。Mainly for the noise in image, illumination is not enough, size and angle is bad etc. reason causes gather after the not high problem of picture quality, adopt image enhaucament means carry out extracting to facial image feature and select。

4) carrying out recognition of face, be calculated processing by the face feature vector of acquisition and the characteristic vector in face database, find out the most close vector, obtain the maximum of result of calculation, being considered as it is the face to look for。

5) result that recognition of face goes out is authorized user's face, voice announcer can report out " identifying successfully " voice messaging, it is the high level of 1 that direct current electromagnetic relay control circuit will be exported signal by face recognition module, audion and field effect transistor in energy supply control module turn on, the control circuit of output 24V DC voltage supply direct current electromagnetic relay, after direct current electromagnetic relay energising, armature is inhaled by electric magnet, movable contact contacts with stationary contact simultaneously, and the electrical equipment of peripheral circuit just can authorized user start。

6) if the face that recognition result is undelegated user, voice announcer can report out " identifying mistake " voice messaging, it is the low level of 0 that direct current electromagnetic relay control circuit can be exported signal by face recognition module, audion is in cut-off state, the grid of field effect transistor and source electrode both end voltage are 0, field effect transistor is in cut-off state, and the control circuit of electromagnetic relay does not turn on, and the electrical equipment of peripheral circuit can not authorized user be started。

Compared with existing system, implementing the equipment used in the present invention less, simple in construction, with low cost, feasibility and the safety of method are higher。The combination of face recognition technology and direct current electromagnetic relay can be made full use of, the effective special equipment controlled in electrical system or high-risk equipment, avoid the improper operation of disabled user, for industrial implementation intellectuality, there is long-range meaning, face recognition technology is adopted to use user to carry out identification electrical equipment, giving the control loop that user identity authority remotely controls direct current electromagnetic relay, controlling thus reaching authorized user's operation to special equipment or high-risk equipment。

Claims (5)

1. the direct current electromagnetic relay device controlled based on recognition of face, it is characterized in that, including face recognition module (1), power supply switch controller (2), voice announcer (3), direct current electromagnetic relay (4), liquid crystal display (5)

Described face recognition module (1) respectively with liquid crystal display (5), on and off switch controls module (2) and voice announcer (3) connects, described on and off switch controls module (2) and is connected with the low-voltage control circuit of direct current electromagnetic relay (4), the facial image of user can be demonstrated on liquid crystal display (5) in real time when face recognition module (1) recognizes user's face, if face recognition module (1) identifies that after successfully, on and off switch control module (2) can export the low-voltage control circuit of 24V DC voltage supply direct current electromagnetic relay (4)。

2. the control method of the direct current electromagnetic relay device controlled based on recognition of face of a claim 1, it is characterised in that specifically implement according to following steps:

Step 1, carries out pretreatment to the facial image collected, and adopts Haar wavelet basis function to calculate rectangular characteristic the structural classification device of image, face is detected,

Step 2, after eigenface vector R (D) pretreatment that strong classifier is exported, characteristic vector R (D) is carried out similar face coupling by recycling AngleContex matrix, improve recognition of face precision, set up the face database of authorized user in advance, then the face characteristic extracted utilize AngleContex matrix be calculated it and the similarity authorizing face database set up, it may be judged whether for authorized user's face。

3. the control method of the direct current electromagnetic relay device controlled based on recognition of face according to claim 2, it is characterised in that comprising the concrete steps that of described step 1:

Step 1.1, rectangular characteristic value is all grey scale pixel value sums and the difference of all grey scale pixel value sums in white portion in black region, have employed a kind of rectangular characteristic being similar to Haar wavelet basis function, rectangular characteristic value in detection window, rectangular characteristic value can be calculated by the gray value of integral image pixel, in integral image, coordinate is that the gray value of (X, Y) pixel is equal to all grey scale pixel values in correspondence position upper left side cumulative on original image；

The computing formula of integral image is as follows:

$I(X,Y)=\underset{x\≤X}{\Σ}\underset{y\≤Y}{\Σ}i(x,y)---(1-1)$

I (X, Y) is the image after integration, i (x, y) for original image,

Step 1.2, trying to achieve the integral image I (X, Y) of pixel as a sequence, forms the integrogram matrix of this figure, finds the set D of matrix exgenvalue according to integrogram matrix,

D={a₁,a₂,a₃,a₄,........a_m}(1-2)

Step 1.3, in rectangular window, builds multistage classifier and obtains the classification function of optimal threshold, matrix exgenvalue is carried out best classification,

Step 1.4, utilizes binarization of gray value and integral projection method to carry out eye location in the human face region detected；With eyes coordinates for datum mark, two eyes are made to be in same level line by translating, rotating, by scale transformation, facial image is carried out size to unify, facial image is carried out grayscale normalization, average and the mean square deviation of facial contour inner region gray scale is calculated by histogram equalization, image smoothing, gray scale normalization etc., view picture facial image is carried out greyscale transformation, highlights face key facial features。

4. the control method of the direct current electromagnetic relay device controlled based on recognition of face according to claim 3, it is characterised in that concretely comprising the following steps of described step 1.3:

Step 1.3.1, constructs training sample set L:

L={ (D₁,K₁),(D₂,K₂),......(D_m,K_m)}(1-3)

Wherein D represents the matrix characteristic vector of sample, and K represents training result, and m is training total number of samples, m=1,2,3 ... .n；

Step 1.3.2, the initial probability distribution of sample L is sample weights:

N represents number of training, and the iterations of grader is t=1,2,3,4.......T, and T is any positive integer, calculates the weights of normalization sample:

$q_{t,m}=\frac{W_{t,m}}{\underset{j=1}{\Σ}{W}_{t,m}}---(1-4)$

Wherein j=1,2,3 ... .n；

Step 1.3.3, calls weak learning algorithm to the matrix characteristic vector D of each sample, trains a Weak Classifier H_j(D_m), and calculate the error rate ξ of this grader_j

$H_j\left(D_m\right)=\left\{\begin{array}{cc}1;& p_j{f}_j\≥p_j{\mathrm{\θ}}_j\\ 0;& otherwise\end{array}\right.---(1-5)$

Wherein θ_jIt is the threshold value of Weak Classifier, p_jFor the biasing of grader, the direction of majorization inequality, f_jFor the eigenvalue of eigenmatrix,

Step 1.3.4, selects have minimal error rate ξ_tCorresponding Weak Classifier H_t(D_m) as optimal classification device, namely

ξ_t=min ξ_j(1-7)

Step 1.3.5, readjusting sample weights according to optimal classification device is:

$W_{t+1,m}=W_{t,m}{{\mathrm{\β}}_t}^{1-e_m}---(1-8)$

Otherwise wherein m-th sample correct judgment then em=0 em=1. so can increase grader error sample weight, reducing the weight of correct sample of classifying, during next iteration, the selection of Weak Classifier will pay attention to the sample of last mistake classification error,

${\mathrm{\β}}_t=\frac{{\mathrm{\ξ}}_t}{1-{\mathrm{\ξ}}_t}---(1-9)$ β_tThe coefficient of sample weights is readjusted for optimal classification device；

Step 1.3.6, generates strong classifier:

$R\left(D\right)=\underset{t=1}{\overset{T}{\Σ}}log\frac{1}{{\mathrm{\β}}_t}{H}_t\left(D_m\right)---(1-10)$

IfThen grader output result is 1, then it is assumed that this detection window is likely human face region, and R (D) is the eigenface vector of output；

The multistage classifier utilizing structure can get rid of most of non-face region, detects almost possible human face region；If every first-level class device is output as 1, then it is assumed that this detection window is likely human face region, the window of detection is input to next stage grader and proceeds to judge, otherwise, in this level, detection window is excluded as non-face region。

5. the control method of direct current electromagnetic relay device controlled based on recognition of face according to right 2, it is characterised in that described step 2 particularly as follows:

When extracting face characteristic, it is identical for belonging to the human face characteristic point distance with a part, and the difference of face is mainly manifested in the difference of face each several part contour line curvature, angle, can with 8 degree of divisions as face characteristic average distance, by 45 regions of plane division being zero with datum mark, generate AngleContex matrix, calculate the number of other discrete points falling into regional

Specifically implement according to following steps:

Step 2.1, on the eigenface that strong classifier obtains, each datum mark is selected for 9 character references points with nose, corners of the mouth midpoint, point, left eyebrow, left eye angle, the left corners of the mouth, right eyebrow, right eye angle, the right corners of the mouth, 45 regions (bin) are chosen with 8 degree of angles for dividing interval, thus can obtain 9*45 AngleContex matrix

Step 2.2, utilizes angular histogram to intersect method and the similarity of 9*45 the AngleContex matrix calculus face to be matched obtained and targeted customer's face, and computing formula is as follows:

$P(s,v)=\underset{l,g}{\Σ}min\[R_{l,g}\left(D\right),{R^{\′}}_{l,g}\left(D\right)\]---(2-1)$

$M(R,R^{\′})=\frac{P(s,v)}{9*45}---(2-2)$

P (s, v) is the crossing distance of two characteristic vectors, and M (R, R') is the ratio of 9*45 overall area (bin) pixel shared by each zoning (bin) in two rectangular histograms,

Wherein, R_l,gRepresent the AngleContex matrix of face to be matched, R'_l,gThe AngleContex matrix of expression targeted customer's face, l=1,2 ..., 9 represent 9 character references points；G=1,2,3,4,5,6 ..., 44,45 expressions choose 45 regions with 8 degree of angles for dividing interval；

Step 2.3, the value calculating M (R, R') between 0～1, can take M (R, R') maximum just for the similar face that matching degree is the highest,

Step 2.4, after when similar face, the match is successful, voice announcer 3 can report the voice of " identifying successfully ", and otherwise, voice announcer 3 can report the voice of " identifying mistake ",

Wherein human face similarity degree coupling, on the eigenface image obtained, with nose, corners of the mouth midpoint, point, left eyebrow, left eye angle, the left corners of the mouth, right eyebrow, right eye angle, the right corners of the mouth for 9 character references points, each datum mark selected on the eigenface image obtained through strong classifier, 45 regions (bin) are chosen with 8 degree of angles for dividing interval, thus can obtaining 9*45 AngleContex matrix, multizone divides and improves matching precision as far as possible。