CN101980243A - Binocular vision-based finger vein three-dimensional identification method and device - Google Patents

Abstract

The invention discloses a binocular vision-based finger vein three-dimensional identification method and a binocular vision-based finger vein three-dimensional identification device. The device mainly comprises an infrared light source, a light touch switch, a left camera, a right camera, infrared filters, a power supply, an advanced RISC machines (ARM) interface board and a digital signal processing (DSP) plate, wherein the infrared light source, a finger and the left camera and the right camera are positioned on three horizontal planes; the left camera and the right camera are positioned on the two sides of a finger projection line respectively; and the connecting line of the left camera and the right camera is vertical to the finger projection line. The method comprises the following steps of: if that the light touch switch is pressed is detected, operating the left camera and the right camera to acquire the binocular image of the finger, and extracting and matching the three-dimensional characteristics of finger veins based on the binocular vision principle and a scale invariant feature transform (SIFT) operator so as to realize the function of registering or identifying the finger. The method and the device effectively improve the identification rate and security level of finger vein identification equipment.

Description

Finger vena three-dimensional recognition method and device based on binocular vision

Technical field

The present invention relates generally to a kind of finger vena three-dimensional recognition method and device based on binocular vision.

Background technology

Finger vena identification is biometrics identification technology of new generation, and its technical advantage mainly contains: the first, and vein identification safe class height, vein is concealed in body interior, transreplication or impaired not, and have only live body that vein is just arranged, its feature is difficult for stealing; The second, the vein pattern individual difference is big, even the similar twinborn vein of looks is also variant, and this species diversity can not disappear in life at them; The 3rd, vein identification is adopted based on ultrared Noninvasive and untouchable imaging technique, easy to use, cleaning; The 4th, it is wide that vein identification is suitable for the crowd, and everybody has vein, can extract vein pattern.Based on above advantage, the application of vein identification technology also more and more is subjected to the attention of enterprise, research institution and government.Existing vein recognition system mainly contains the middle finger vena identification system of HIT, the palm vein Verification System of Fujitsu and the hand back vein Verification System of Singapore VEID company; Tokyo Mitsubishi Bank will have released vena identification withdraw the money card and ATM (automatic teller machine) in the end of the year 2004, be subjected to user's extensive welcome.Up to the present, vein recognition system has been successfully applied to fields such as gate inhibition, ATM automatic teller machine, note-book computer lock.

Yet, the finger vena plane projection image that existing finger vein identification method all only arrives at the single image sensor acquisition launches research, because losing of depth information, in case before and after same finger during twice imaging with camera lens between distance different, or self attitude changes, and the vein image that will cause collecting exists than big-difference.For guaranteeing recognition performance, existing equipment all requires the disposing way of the each finger of user to be consistent in actual applications as far as possible, brings greater inconvenience to the user.On the other hand, for the two cover collecting devices that the collection structure there are differences, even identical finger and the identical posture of putting, the vein image that is obtained still there are differences, the versatility that is the vein image feature that gets access to of distinct device is relatively poor, recognition performance descends seriously when some registration, multiple spot identification, has limited the usable range of equipment.Simultaneously, the high security that existing vein identification equipment is declared is based on everyone vein distribution this fact all inequality, but the uniqueness of three-D space structure can not guarantee its two-dimensional projection and also have uniqueness, the different corresponding similar two-dimensional projections of three-dimensional structure possibility, the obtaining and imitated do not need very high cost of two-dimensional projection in addition, therefore existing vein identification equipment and be not suitable for the exigent application places of security performance.

The fundamental way that addresses the above problem is to adopt three-dimensional identification to replace two dimension identification.When finger stretched, the three-D space structure of finger vena was approximately a space pipe network structure, when the putting position of finger and the anglec of rotation when changing, did not influence the pipe network inner structure.Therefore, three-dimensional identification can overcome the adverse effect of finger gesture variation to recognition performance effectively, the versatility of the vein image feature that the raising distinct device gets access to.Simultaneously because the uniqueness of finger vena three-D space structure and the imitated difficulty of three-dimensional feature make three-dimensional vein identification equipment will possess higher security.

Summary of the invention

For solving the problem of existing vein identification, the invention provides a kind of finger vena three-dimensional recognition method and device based on binocular vision, utilize the principle of binocular vision to build finger vena binocular acquisition platform, obtain the three-dimensional information of finger venous image in conjunction with SIFT operator and binocular vision principle, in conjunction with the three-dimensional identification of SIFT feature and space length realization finger vena, improve the discrimination and the level of security of finger vena identification equipment effectively.

Device comprise infrared light supply, touch-switch, about two video cameras, infrared filter, power supply, ARM interface board and DSP disposable plates, two parallel placements of video camera about it is characterized in that, infrared light supply, finger, about two video cameras on three surface levels, about two video cameras lay respectively at the both sides of finger projection line, about the line of two video cameras vertical with the finger projection line, and can divide the finger projection line equally substantially.

Employing is obtained the three-dimensional information of finger venous image based on SIFT operator and binocular vision, extracts the three-dimensional feature of finger vena; Employing realizes the three-dimensional feature identification of finger vena based on the method for SIFT feature and space length.

Concrete technical scheme is:

One, finger vena binocular acquisition platform design

Based on the principle of binocular vision, the present invention designs the binocular acquisition platform, is used to obtain the three-dimensional feature of finger vena, and wherein, Fig. 1 is the structural drawing of binocular acquisition platform, and Fig. 2 is the circuit connection diagram of binocular acquisition platform.The binocular acquisition platform mainly comprise infrared light supply, touch-switch, about two video cameras, infrared filter, power supply, ARM interface board, DSP disposable plates, wherein infrared light supply, finger, about two video cameras on three surface levels, about two video cameras lay respectively at the both sides of finger projection line, about the line of two video cameras with to point projection line vertical.

On structural design, following principle is arranged specifically:

The first, infrared light supply, finger, video camera three are on three surface levels, and the about 1cm of the vertical range of infrared light supply and finger, the about 6cm of the vertical range of video camera and finger;

The second, left and right cameras be positioned at the finger projection line both sides, and with the finger projection line horizontal range all be 3cm;

The 3rd, about the line of two video cameras vertical with the finger projection line, and can divide the finger projection line equally substantially.

The 4th, for fear of the interference of external light source, on the camera lens of left and right cameras, all added infrared filter, the non-infrared light of filtering.

The circuit connecting relation of binocular acquisition platform as shown in Figure 2, concrete annexation is:

Power module is mainly used in to infrared light supply, video camera, ARM interface board, the power supply of DSP disposable plates; The ARM interface board is mainly realized human-computer interaction function, receives the command information from the user, communicates with the DSP disposable plates then, realizes function corresponding; The DSP disposable plates at first receives the control command of ARM plate, detect touch-switch then, to judge whether to exist pending finger, if detecting touch-switch presses, then operate left and right cameras collection finger binocular image, the method of line correlation of going forward side by side is handled, and realizes the function of finger login or identification.

Two, based on the finger venous image 3 D information obtaining method of SIFT operator and binocular vision

The ultimate principle of binocular vision is: at first to left and right camera calibration, determine image transformation matrix separately; In left camera review, specify a pixel A then, in right camera review, carry out solid coupling, find out the pixel B that matches, obtain three-dimensional put to (A, B); Ask the intersecting point coordinate of space ray OA and OB at last, be three-dimensional point (A, B) common corresponding volume coordinate.

The key of utilizing binocular vision to carry out three-dimensional reconstruction is that how to obtain abundant and reliable and stable solid point right.The SIFT operator is that a kind of image local feature that maintains the invariance based on metric space, to image zoom, rotation even affined transformation is described operator, and operator stability and uniqueness are all relatively good.

The present invention adopts the finger venous image 3 D information obtaining method based on SIFT operator and binocular vision, at first adopt based on the method for 3D target to about two video cameras calibrate, obtain two video cameras image transformation matrix separately; The key point of two width of cloth finger venous images about adopting the SIFT operator to obtain respectively then, and extract the SIFT proper vector of each key point; Then adopt the characteristic matching strategy based on SIFT to carry out the solid coupling, the space multistory point that is met the SIFT matching condition is right; At last according to the ultimate principle of binocular vision, with space multistory put right planimetric coordinates respectively the substitution correspondence about the image transformation matrix of two video cameras, ask for the right volume coordinate of all three-dimensional points.The overall flow of method as shown in Figure 3.

Wherein, the present invention adopts SIFT operator extraction key point feature and realizes three-dimensional coupling, and it is right to obtain reliable and stable space multistory point.When obtaining the right volume coordinate of each three-dimensional point, adopt the method for look-up table to ask for the right volume coordinate of three-dimensional point fast.

The concrete steps of this method are as follows:

(1) camera calibration

Camera calibration is a numerical process determining the video camera projection matrix.If inpolar O ₁In that (u, v) the coordinate in the coordinate system is (u ₀, v ₀), its three dimensional space coordinate be (x, y, z), image coordinate be (u, v), the equation below the corresponding relation of image coordinate and volume coordinate satisfies:

$z_c\left[\begin{array}{c}u\\ v\\ 1\end{array}\right]=\begin{array}{}\end{array}\left[\begin{array}{cccc}f/\mathrm{dx}& 0& u_0& 0\\ 0& f/\mathrm{dy}& v_0& 0\\ 0& 0& 1& 0\end{array}\right]\left[\begin{array}{cc}R& t\\ 0& 1\end{array}\right]\left[\begin{array}{c}x\\ y\\ z\\ 1\end{array}\right]$

Wherein, dx, dy represent the physical size of single pixel at x axle and y direction of principal axis entropy respectively, and f represents focal length of camera, and these all are the inner parameters of video camera; R is quadrature unit's rotation matrix of 3 * 3, and t is that 3 * 1 translation is adjacent, and the two is used for describing the position and the angular deflection of camera coordinate system and world coordinate system, is the external parameter of video camera.Z _cBe image slices vegetarian refreshments to be asked (u, depth coordinate v).For convenience, can note by abridging and be:

$z_c\left[\begin{array}{c}u\\ v\\ 1\end{array}\right]=\left[\begin{array}{cccc}{m}_{11}& m_{12}& m_{13}& m_{14}\\ m_{21}& m_{22}& m_{23}& m_{24}\\ m_{31}& m_{32}& m_{33}& m_{34}\end{array}\right]\left[\begin{array}{c}x\\ y\\ z\\ 1\end{array}\right]$

The purpose of camera calibration is to ask for the inside and outside parameter of video camera.Thereby, can extrapolate its volume coordinate for the arbitrary pixel coordinate on the image.The method of camera calibration has a lot, because finger vena harvester of the present invention video camera installation site and fixed angle, can adopt method to realize camera calibration based on the 3D target, concrete grammar is: the 3D target easily asked for of selected characteristic point (as the three-dimensional inclined-plane that is become by chequered with black and white combinations of blocks) at first, the manual three-dimensional coordinate of measuring certain characteristics point; Calculate the pixel coordinate of these these unique points of unique point labor measurement in image then, the substitution equation utilizes each unknown quantity in the least square method solving equation, thereby finishes camera calibration.

After camera calibration finishes, get final product images acquired, gather respectively parallel placement about two camera reviews, constitute the binocular image.

(2) extract the image key points feature

Step1: adopt difference of Gaussian (DoG) filtering method that image is carried out multi-scale filtering, obtain the DoG metric space.The DoG operator is described as with formula:

D(x，y，σ)＝(G(x，y，kσ)-G(x，y，σ))*I(x，y)

Wherein, σ represents the variance of Gaussian function, I (x, y) the remarked pixel point (x, the gray-scale value of y) locating, k are represented the number of plies of yardstick, the formula of G correspondence is:

$G(x,y,\mathrm{\σ})=\frac{1}{2\mathrm{\π}{\mathrm{\σ}}^2}{e}^{-(x^2+y^2){2\mathrm{\σ}}^2}$

Setp2: the Local Extremum of detected image in the DoG metric space, reject the Local Extremum of the low and edge of contrast, with the Local Extremum of remainder as key point.

Setp3: the direction character that extracts key point, binding site and scale feature generate the SIFT proper vector again, then proper vector are carried out normalization, remove the influence of rotation, dimensional variation and illumination variation, this proper vector is the proper vector of finger venous image key point.In the experiment each key point adopt 4 * 4 totally 16 seed points be described, each seed points comprises the information of 8 directions, such key point will produce the SIFT proper vector of 16 * 8=128 dimension.

(3) three-dimensional coupling

After the key point proper vector of binocular image generates, adopt characteristic matching strategy to carry out the solid coupling based on SIFT.Three-dimensional matching strategy is: get the some key points in the left image, and find out European nearest preceding two key points in itself and the right image, in these two key points, if nearest distance is removed near in proper order distance less than certain proportion threshold value (proportion threshold value is made as 80% in the experiment), then accept this a pair of match point, one group of space multistory point that this a pair of match point is in the binocular image is right; Otherwise, reject this key point.In this way, travel through the key point on the left image, the space multistory point that obtains all couplings is right.

(4) calculate three-dimensional point to volume coordinate

According to the ultimate principle of binocular vision, space multistory is put the right planimetric coordinates left and right video camera imaging transformation matrix of substitution correspondence respectively, the intersecting point coordinate of asking for two space raies is the right volume coordinate of three-dimensional point, and concrete grammar can be described below:

If a certain group of space multistory point is to being (P ₁, P ₂), its planimetric coordinates is respectively (u ₁, v ₁), (u ₂, v ₂), corresponding depth coordinate is respectively Z _C1, Z _C2Three-dimensional point is to corresponding to the same point on the space, remember its volume coordinate for (x, y, z).Image transformation matrix with planimetric coordinates and volume coordinate difference substitution left and right cameras obtains two following equations:

$z_{c1}\left[\begin{array}{c}u1\\ v1\\ 1\end{array}\right]=\left[\begin{array}{cccc}{m}_{11}& m_{12}& m_{13}& m_{14}\\ m_{21}& m_{22}& m_{23}& m_{24}\\ m_{31}& m_{32}& m_{33}& m_{34}\end{array}\right]\left[\begin{array}{c}x\\ y\\ z\\ 1\end{array}\right]$

$z_{c2}\left[\begin{array}{c}u2\\ v2\\ 1\end{array}\right]=\left[\begin{array}{cccc}{M}_{11}& M_{12}& M_{13}& M_{14}\\ M_{21}& M_{22}& M_{23}& M_{24}\\ M_{31}& M_{32}& M_{33}& M_{34}\end{array}\right]\left[\begin{array}{c}x\\ y\\ z\\ 1\end{array}\right]$

The group of solving an equation obtains three-dimensional point to (P ₁, P ₂) volume coordinate (x, y, z).

By that analogy, obtain the right volume coordinate of all three-dimensional points.

In order to improve arithmetic speed, can set up look-up table, make (x, y, z)=f (u ₁, v ₁, u ₂, v ₂), according to the image transformation matrix of left and right cameras, (x, y z), just can directly search the volume coordinate of Different Plane coordinate correspondence to the volume coordinate of calculated in advance planimetric coordinates correspondence like this in subsequent process.

Three, based on the finger vena three-dimensional feature recognition methods of SIFT feature and space length

Finger vena three-dimensional feature recognition methods based on SIFT feature and space length is: under logging status, obtain the SIFT proper vector of each key point of finger vena and the three-dimensional feature that volume coordinate is formed, and be stored among the FLASH; Under status recognition, obtain each key point of finger vena three-dimensional feature, be complementary with the finger vena three-dimensional feature that is stored among the FLASH then, whether differentiate finger according to matching result legal.

The three-dimensional feature of finger vena comprises two parts, the one, and the plane characteristic of each key point also is its SIFT proper vector, describes the plane projection characteristic of finger vena with it; The 2nd, the stereoscopic features of each key point also is its volume coordinate, describes the three-dimensional distribution character of finger vena with it.These features were all obtained in the Depth Information Acquistion stage, only needed here with specific data structure its storage to be got final product.

The matching criterior of three-dimensional feature is:

Step1: adopt SIFT characteristic matching strategy, the SIFT feature of SIFT feature in the property data base and vein image to be identified is mated, the key point that is met the SIFT matching condition is right.Wherein, because the SIFT feature has two groups: left image SIFT feature and right image SIFT feature, so the SIFT characteristic matching here requires the left characteristics of image of finger to be identified and the left characteristics of image in the database to mate, the right characteristics of image of finger to be identified and the right characteristics of image in the database mate;

Step2: ask for the space length d of the right spatial coordinate of key point,

$d=\sqrt{{(x1-x2)}^2+{(y1-y2)}^2+{(z1-z2)}^2}$

If distance is greater than a certain threshold value D, it is right then to reject this group key point.

Step3: calculate the remaining right number of key point, and,, think that then two finger venous images are similar if should count greater than a certain threshold value N with its similarity determination tolerance as two finger venous images; Otherwise, think two finger venous image dissmilarities.

Wherein, by a large amount of experiment statisticses, choose D=5, N=7.

The invention has the advantages that the discrimination and the level of security that improve the finger vena identification equipment effectively.

Description of drawings

Fig. 1 is a finger vena binocular acquisition platform structural drawing;

Fig. 2 is a finger vena binocular acquisition platform circuit connection diagram;

Fig. 3 is a finger venous image 3 D information obtaining method process flow diagram;

Fig. 4 is that the finger vena three-dimensional feature extracts and the coupling process flow diagram.

Embodiment

Fig. 1 is a finger vena binocular acquisition platform structural drawing.The binocular acquisition platform mainly comprise infrared light supply, touch-switch, about two video cameras, infrared filter, power supply, ARM interface board, DSP disposable plates, wherein infrared light supply, finger, about two video cameras on three surface levels, about two video cameras lay respectively at the both sides of finger projection line, about the line of two video cameras vertical with the finger projection line, and can divide the finger projection line equally substantially.All added infrared filter on the camera lens of left and right cameras, the non-infrared light of filtering.

Fig. 2 is a finger vena binocular acquisition platform circuit connection diagram.Power module is mainly used in to infrared light supply, video camera, ARM interface board, the power supply of DSP disposable plates; The ARM interface board is mainly realized human-computer interaction function, receives the command information from the user, communicates with the DSP disposable plates then, realizes function corresponding; The DSP disposable plates at first receives the control command of ARM plate, detect touch-switch then, to judge whether to exist pending finger, if detecting touch-switch presses, then operate left and right cameras collection finger binocular image, the method of line correlation of going forward side by side is handled, and realizes the function of finger login or identification.

Fig. 3 is a finger venous image 3 D information obtaining method process flow diagram.At first, employing based on the method for 3D target to about two video cameras calibrate, the inside and outside parameter of calibrating camera, obtain two video cameras image transformation matrix separately, after camera calibration finishes, get final product images acquired, gather respectively parallel placement about two camera reviews, constitute the binocular image; Then, adopt the SIFT operator to obtain respectively about the key point of two width of cloth finger venous images, and extract the SIFT proper vector of each key point; Then, adopt the characteristic matching strategy based on SIFT to carry out the solid coupling, the space multistory point that is met the SIFT matching condition is right; At last,, space multistory is put the right planimetric coordinates left and right video camera imaging transformation matrix of substitution correspondence respectively, ask for the right volume coordinate of all three-dimensional points according to the ultimate principle of binocular vision.

Fig. 4 is that the finger vena three-dimensional feature extracts and the coupling process flow diagram.Finger vein recognition system has login and discerns two states, under logging status, adopts the finger venous image 3 D information obtaining method based on SIFT and binocular vision, obtains the three-dimensional feature of finger vena and is stored among the FLASH; Under status recognition, adopting uses the same method obtains the three-dimensional feature of finger vena, is complementary with the finger vena three-dimensional feature that is stored among the FLASH then, judges according to matching criterior whether this finger is legal.The three-dimensional feature of finger vena comprises two parts, the one, and the plane characteristic of each key point also is its SIFT proper vector, describes the plane projection characteristic of finger vena with it; The 2nd, the stereoscopic features of each key point also is its volume coordinate, describes the three-dimensional distribution character of finger vena with it.These features were all obtained in the Depth Information Acquistion stage, only needed here with specific data structure its storage to be got final product.The matching criterior of three-dimensional feature is: at first, adopt SIFT characteristic matching strategy, the SIFT feature of SIFT feature in the property data base and vein image to be identified is mated, the key point that is met the SIFT matching condition is right.Wherein, because the SIFT feature has two groups: left image SIFT feature and right image SIFT feature, so the SIFT characteristic matching here requires the left characteristics of image of finger to be identified and the left characteristics of image in the database to mate, the right characteristics of image of finger to be identified and the right characteristics of image in the database mate; Then, ask for the space length of the right spatial coordinate of key point, if the distance greater than a certain threshold value, it is right then to reject this group key point; At last, calculate the right number of key point of remainder, and with its similarity determination tolerance as two finger venous images, if should count greater than a certain threshold value, think that then two finger venous images are similar, it is legal promptly should to point; Otherwise, think two finger venous image dissmilarities, promptly should finger illegal.

Claims (7)

1. based on the finger vena three-dimensional recognition method and the device of binocular vision, it is characterized in that device comprise infrared light supply, touch-switch, about two video cameras, infrared filter, power supply, ARM interface board and DSP disposable plates, two parallel placements of video camera about it is characterized in that, infrared light supply, finger, about two video cameras on three surface levels, about two video cameras lay respectively at the both sides of finger projection line, about the line of two video cameras vertical with the finger projection line, and can divide the finger projection line equally substantially.

2. based on the finger vena three-dimensional recognition method of binocular vision, it is characterized in that adopting the three-dimensional information that obtains finger venous image based on SIFT operator and binocular vision, extract the three-dimensional feature of finger vena; Employing realizes the three-dimensional feature identification of finger vena based on the method for SIFT feature and space length.

3. the finger vena three-dimensional recognition method based on binocular vision according to claim 2, it is characterized in that based on the finger venous image 3 D information obtaining method of SIFT operator and binocular vision being: at first adopt based on the method for 3D target to about two video cameras calibrate, obtain two video cameras image transformation matrix separately; The key point of two width of cloth finger venous images about adopting the SIFT operator to obtain respectively then, and extract the SIFT proper vector of each key point; Then adopt the characteristic matching strategy based on SIFT to carry out the solid coupling, the space multistory point that is met the SIFT matching condition is right; At last according to the ultimate principle of binocular vision, with space multistory put right planimetric coordinates respectively the substitution correspondence about the image transformation matrix of two video cameras, ask for the right volume coordinate of all three-dimensional points.

4. the finger vena three-dimensional recognition method based on binocular vision according to claim 2 is characterized in that adopting SIFT operator extraction key point feature and realizes three-dimensional coupling, and it is right to obtain reliable and stable space multistory point.

5. the finger vena three-dimensional recognition method based on binocular vision according to claim 2 is characterized in that adopting the method for look-up table to ask for the right volume coordinate of three-dimensional point fast.

6. the finger vena three-dimensional recognition method based on binocular vision according to claim 2, it is characterized in that based on the finger vena three-dimensional feature recognition methods of SIFT feature and space length being: under logging status, obtain the SIFT proper vector of each key point of finger vena and the three-dimensional feature that volume coordinate is formed, and be stored among the FLASH; Under status recognition, obtain the three-dimensional feature of each key point of finger vena, be complementary with the finger vena three-dimensional feature that is stored among the FLASH then, whether differentiate finger according to matching result legal.

7. the finger vena three-dimensional recognition method based on binocular vision according to claim 2, the matching criterior that it is characterized in that three-dimensional feature is: at first adopt SIFT characteristic matching strategy, the SIFT feature of SIFT feature in the property data base and vein image to be identified is mated, and the key point that is met the SIFT matching condition is right; Ask for the space length of the right spatial coordinate of key point then, if the distance greater than a certain threshold value, it is right then to reject this group key point; Calculate the right number of key point of remainder at last, and with its similarity determination tolerance as two finger venous images, if should count greater than a certain threshold value, think that then two finger venous images are similar, it is legal promptly should to point; Otherwise, think two finger venous image dissmilarities, promptly should finger illegal.

Images